Documentation 2026-09-21 23:02 UTC to 2026-09-22 23:00 UTC

19GPT-5.6 and GPT-5.5 consume credits at 2.5x the Standard rate; GPT-5.4 consumes19GPT-5.6 and GPT-5.5 consume credits at 2.5x the Standard rate; GPT-5.4 consumes

20credits at 2x the Standard rate.20credits at 2x the Standard rate.

21 21 

22GPT-6 Astra Fast mode consumes credits at 2.5x the Standard rate where22For GPT-6 Astra, GPT-6 Sol, and GPT-6 Luna, Fast mode consumes credits at

23available. See [Models](https://learn.chatgpt.com/docs/models) for model availability and232.5x the Standard rate where available. See [Models](https://learn.chatgpt.com/docs/models) for model availability and

24[Pricing](https://learn.chatgpt.com/docs/pricing#token-rates) for token rates.24[Pricing](https://learn.chatgpt.com/docs/pricing#token-rates) for token rates.

25 25 

26Use `/fast on`, `/fast off`, or `/fast status` in the CLI to change or inspect26Use `/fast on`, `/fast off`, or `/fast status` in the CLI to change or inspect

162request a specific model or reasoning effort in your prompt,162request a specific model or reasoning effort in your prompt,

163configure `[agents]` defaults in `config.toml`, or set `model` and163configure `[agents]` defaults in `config.toml`, or set `model` and

164`model_reasoning_effort` directly in the custom agent file.164`model_reasoning_effort` directly in the custom agent file.

165For example, use `gpt-5.6-terra` for fast scans or a higher-effort `gpt-5.6` configuration for more demanding reasoning.165For example, use `gpt-6-luna` for fast scans or a higher-effort `gpt-6-sol` configuration for more demanding reasoning.

166 166 

167For most tasks in Codex, start with 167For most tasks in Codex, start with

168 `gpt-5.6`. Use 168 `gpt-6-sol`. Use

169 `gpt-5.6-terra` when you want169 `gpt-6-luna` when you want a faster, lower-cost option for lighter subagent

170 a faster, lower-cost option for lighter subagent work.170 work.

171 171 

172### Model choice172### Model choice

173 173 

174- **`gpt-5.6`**: Start here for demanding agents. It's strongest for ambiguous, multi-step work that needs planning, tool use, validation, and follow-through across a larger context.174- **`gpt-6-sol`**: Start here for demanding agents. It's strongest for ambiguous, multi-step work that needs planning, tool use, validation, and follow-through across a larger context.

175- **`gpt-5.6-terra`**: Use for agents that favor speed and efficiency over depth, such as exploration, read-heavy scans, large-file review, or processing supporting documents. It works well for parallel workers that return distilled results to the main agent.175- **`gpt-6-luna`**: Use for fast, narrowly scoped agents handling clear, repeatable, or high-volume work.

176- **`gpt-5.6-luna`**: Use for fast, narrowly scoped agents handling clear, repeatable, or high-volume work.

177 176 

178### Reasoning effort (`model_reasoning_effort`)177### Reasoning effort (`model_reasoning_effort`)

179 178 

179For explicit model settings, start with `medium` for GPT-6 Sol, `high` for

180GPT-6 Luna, or `low` for GPT-6 Astra. Adjust for the task using a level the

181selected model supports.

182 

180- **`ultra`**: Use for the deepest reasoning when the selected model supports183- **`ultra`**: Use for the deepest reasoning when the selected model supports

181 it.184 it.

182- **`max`** and **`xhigh`**: Use for especially demanding reasoning when the185- **`max`** and **`xhigh`**: Use for especially demanding reasoning when the

183 selected model supports these levels.186 selected model supports these levels.

184- **`high`**: Use when an agent needs to trace complex logic, check assumptions, or work through edge cases (for example, reviewer or security-focused agents).187- **`high`**: Use when an agent needs to trace complex logic, check assumptions, or work through edge cases (for example, reviewer or security-focused agents).

185- **`medium`**: A balanced default for most agents.188- **`medium`**: Balances speed and depth; the starting point for GPT-6 Sol.

186- **`low`**: Use when the task is straightforward and speed matters most.189- **`low`**: Use when the task is straightforward and speed matters most.

187 190 

188Higher reasoning effort increases response time and token usage, but it can improve quality for complex work. For details, see [Models](https://learn.chatgpt.com/docs/models), [Config basics](https://learn.chatgpt.com/docs/config-file/config-basic), and [Configuration Reference](https://learn.chatgpt.com/docs/config-file/config-reference).191Higher reasoning effort increases response time and token usage, but it can improve quality for complex work. For details, see [Models](https://learn.chatgpt.com/docs/models), [Config basics](https://learn.chatgpt.com/docs/config-file/config-basic), and [Configuration Reference](https://learn.chatgpt.com/docs/config-file/config-reference).

420```toml423```toml

421name = "pr_explorer"424name = "pr_explorer"

422description = "Read-only codebase explorer for gathering evidence before changes are proposed."425description = "Read-only codebase explorer for gathering evidence before changes are proposed."

423model = "gpt-5.6-luna"426model = "gpt-6-luna"

424model_reasoning_effort = "medium"427model_reasoning_effort = "high"

425sandbox_mode = "read-only"428sandbox_mode = "read-only"

426developer_instructions = """429developer_instructions = """

427Stay in exploration mode.430Stay in exploration mode.

435```toml438```toml

436name = "reviewer"439name = "reviewer"

437description = "PR reviewer focused on correctness, security, and missing tests."440description = "PR reviewer focused on correctness, security, and missing tests."

438model = "gpt-5.6-terra"441model = "gpt-6-sol"

439model_reasoning_effort = "high"442model_reasoning_effort = "medium"

440sandbox_mode = "read-only"443sandbox_mode = "read-only"

441developer_instructions = """444developer_instructions = """

442Review code like an owner.445Review code like an owner.

450```toml453```toml

451name = "docs_researcher"454name = "docs_researcher"

452description = "Documentation specialist that uses the docs MCP server to verify APIs and framework behavior."455description = "Documentation specialist that uses the docs MCP server to verify APIs and framework behavior."

453model = "gpt-5.6-luna"456model = "gpt-6-luna"

454model_reasoning_effort = "medium"457model_reasoning_effort = "high"

455sandbox_mode = "read-only"458sandbox_mode = "read-only"

456developer_instructions = """459developer_instructions = """

457Use the docs MCP server to confirm APIs, options, and version-specific behavior.460Use the docs MCP server to confirm APIs, options, and version-specific behavior.

485```toml488```toml

486name = "code_mapper"489name = "code_mapper"

487description = "Read-only codebase explorer for locating the relevant frontend and backend code paths."490description = "Read-only codebase explorer for locating the relevant frontend and backend code paths."

488model = "gpt-5.6-luna"491model = "gpt-6-luna"

489model_reasoning_effort = "medium"492model_reasoning_effort = "high"

490sandbox_mode = "read-only"493sandbox_mode = "read-only"

491developer_instructions = """494developer_instructions = """

492Map the code that owns the failing UI flow.495Map the code that owns the failing UI flow.

499```toml502```toml

500name = "browser_debugger"503name = "browser_debugger"

501description = "UI debugger that uses browser tooling to reproduce issues and capture evidence."504description = "UI debugger that uses browser tooling to reproduce issues and capture evidence."

502model = "gpt-5.6-terra"505model = "gpt-6-sol"

503model_reasoning_effort = "high"506model_reasoning_effort = "medium"

504sandbox_mode = "workspace-write"507sandbox_mode = "workspace-write"

505developer_instructions = """508developer_instructions = """

506Reproduce the issue in the browser, capture exact steps, and report what the UI actually does.509Reproduce the issue in the browser, capture exact steps, and report what the UI actually does.

518```toml521```toml

519name = "ui_fixer"522name = "ui_fixer"

520description = "Implementation-focused agent for small, targeted fixes after the issue is understood."523description = "Implementation-focused agent for small, targeted fixes after the issue is understood."

521model = "gpt-5.6-luna"524model = "gpt-6-luna"

522model_reasoning_effort = "medium"525model_reasoning_effort = "high"

523developer_instructions = """526developer_instructions = """

524Own the fix once the issue is reproduced.527Own the fix once the issue is reproduced.

525Make the smallest defensible change, keep unrelated files untouched, and validate only the behavior you changed.528Make the smallest defensible change, keep unrelated files untouched, and validate only the behavior you changed.

21 21 

22Make sure you have:22Make sure you have:

23 23 

24- Credentials for the AWS account you want to use

24- Access to supported OpenAI models in Amazon Bedrock.25- Access to supported OpenAI models in Amazon Bedrock.

25- An AWS Region where the selected model is available.26- Access to an AWS Region where the selected model is available.

26- Authentication for the Amazon Bedrock Mantle path configured for the AWS

27 account.

28 27 

29## Configure the provider28## Configure the provider

30 29 

31Add the `amazon-bedrock` model provider for the Amazon Bedrock Mantle path to30Codex allows you to configure the provider by setting `model_provider` in `~/.codex/config.toml`. The ChatGPT desktop app, Codex CLI, IDE extension, and SDK read the same local configuration layers.

32`~/.codex/config.toml`. The ChatGPT desktop app, Codex CLI, IDE extension, and31 

33SDK read the same local configuration layers. Supplying a model is optional.32Choose the provider for the Amazon Bedrock endpoint you want to use: Bedrock

34Select a supported model explicitly when needed.33Runtime for cross-Region inference (CRIS), or Bedrock Mantle for in-Region inference.

34 

35For the Bedrock Runtime endpoint:

36 

37```toml

38model_provider = "amazon-bedrock-runtime"

39```

40 

41For the Bedrock Mantle endpoint:

35 42 

36```toml43```toml

37model_provider = "amazon-bedrock"44model_provider = "amazon-bedrock"

38```45```

39 46 

40This guide covers the Amazon Bedrock Mantle path in supported commercial AWS47### Choose a model

41 Regions. Local ChatGPT Work and Codex surfaces don't support Bedrock Mantle48 

42 endpoints in AWS GovCloud Regions.49Codex uses the configured `model_provider` to choose which models appear in the

50model picker: models supported through the Bedrock Runtime endpoint for

51`amazon-bedrock-runtime`, or through the Bedrock Mantle endpoint for

52`amazon-bedrock`.

53 

54You can optionally specify a [supported model](#supported-models) in the

55configuration file.

56 

57Model availability varies by AWS Region. Refer to AWS [Regional availability by models](https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html#model-regions-openai). Local ChatGPT Work and Codex surfaces don't support Bedrock Mantle endpoints in AWS GovCloud Regions.

43 58 

44## Authentication options59## Authentication options

45 60 

120 135 

121## Verify setup136## Verify setup

122 137 

123- In Codex CLI, open `/status` and confirm Codex is using the138- In Codex CLI, open `/status` and confirm the model provider matches your

124 `amazon-bedrock` model provider.139 endpoint: `amazon-bedrock` for Mantle in-Region inference, or

140 `amazon-bedrock-runtime` for Runtime Global or Geo cross-Region inference.

125- In the ChatGPT desktop app, select Work or Codex and start a new task after141- In the ChatGPT desktop app, select Work or Codex and start a new task after

126 restarting the app.142 restarting the app.

127- In the IDE extension, start a new session after restarting the extension.143- In the IDE extension, start a new session after restarting the extension.

130 146 

131## Supported models147## Supported models

132 148 

133Use exact model IDs:149Use an inference profile ID for Bedrock Runtime provider or a model ID for Bedrock Mantle provider.

150The selected model or profile must be available in your AWS Region and accessible to your

151AWS identity.

152 

153### Global and Geo cross-Region inference using the Bedrock Runtime endpoint

154 

155Global CRIS can route requests to supported

156commercial AWS Regions worldwide, whereas Geo CRIS routes requests within the profile's geography.

157 

158Use `model_provider = "amazon-bedrock-runtime"` with the optional `model` configuration set to an inference profile ID from the following lists. The provider uses

159`https://bedrock-runtime.{region}.amazonaws.com/openai/v1`, where `{region}` is

160the supported source AWS Region from which you send requests. Both Global and Geo CRIS use this endpoint address.

161 

162#### Global CRIS

134 163 

135```text164Supported models and inference profile IDs:

136openai.gpt-5.6-sol165 

137openai.gpt-5.6-terra166- GPT-6 Astra: `global.openai.gpt-6-astra`

138openai.gpt-5.6-luna167- GPT-5.6 Sol: `global.openai.gpt-5.6-sol`

139openai.gpt-5.5168- GPT-5.6 Terra: `global.openai.gpt-5.6-terra`

140openai.gpt-5.4169- GPT-5.6 Luna: `global.openai.gpt-5.6-luna`

170 

171For example, configure Astra with Global CRIS in `~/.codex/config.toml`:

172 

173```toml

174model_provider = "amazon-bedrock-runtime"

175model = "global.openai.gpt-6-astra"

141```176```

142 177 

143Model availability varies by AWS Region. Before selecting a model, see [model178#### United States Geo CRIS

144support by AWS179 

145Region](https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html).180Supported models and inference profile IDs:

181 

182- GPT-6 Astra: `us.openai.gpt-6-astra`

183- GPT-5.6 Sol: `us.openai.gpt-5.6-sol`

184- GPT-5.6 Terra: `us.openai.gpt-5.6-terra`

185- GPT-5.6 Luna: `us.openai.gpt-5.6-luna`

186 

187Codex's built-in Runtime model picker lists the United States Geo and Global

188variants. AWS also lists India geographic inference profile IDs `in.openai.gpt-5.6-terra` and

189`in.openai.gpt-5.6-luna` for `ap-south-1` (Mumbai) and `ap-south-2` (Hyderabad).

190See the AWS [Terra](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-terra.html) and [Luna](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-luna.html) model pages for details.

191 

192Model and CRIS availability vary by source AWS Region. See AWS [Supported Regions and models for inference profiles](https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-support.html), which links to each model's exact inference profile IDs and regional

193availability, and AWS [Regional availability by models](https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html#model-regions-openai) before selecting a provider and a model.

194 

195### In-Region inference using the Bedrock Mantle endpoint

196 

197Use `model_provider = "amazon-bedrock"` with an optional model ID. The provider uses

198`https://bedrock-mantle.{region}.api.aws/openai/v1`.

199 

200Supported models and model IDs:

201 

202- GPT-6 Astra: `openai.gpt-6-astra`

203- GPT-5.6 Sol: `openai.gpt-5.6-sol`

204- GPT-5.6 Terra: `openai.gpt-5.6-terra`

205- GPT-5.6 Luna: `openai.gpt-5.6-luna`

206- GPT-5.5: `openai.gpt-5.5`

207- GPT-5.4: `openai.gpt-5.4`

208 

209Model availability varies by AWS Region. See AWS [Regional availability by models](https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html#model-regions-openai) before selecting a provider and a model. For GPT-6 Astra, refer to the [Bedrock model page for GPT-6 Astra](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-astra.html).

146 210 

147## Feature availability211## Feature availability

148 212 

630If setup fails, check the following:694If setup fails, check the following:

631 695 

632- The model ID exactly matches a supported model.696- The model ID exactly matches a supported model.

697- You use the correct model provider for the endpoint: `amazon-bedrock-runtime` for Runtime endpoint or `amazon-bedrock` for Mantle endpoint.

633- You specify an AWS Region where the model is available.698- You specify an AWS Region where the model is available.

634- The Bedrock API key or AWS credentials are valid and not expired.699- The Bedrock API key or AWS credentials are valid and not expired.

635- The AWS identity has permission to access the selected Bedrock model.700- The AWS identity has permission to access the selected Bedrock model.

108Requests include `method`, `params`, and `id`:108Requests include `method`, `params`, and `id`:

109 109 

110```json110```json

111{ "method": "thread/start", "id": 10, "params": { "model": "gpt-5.6-terra" } }111{ "method": "thread/start", "id": 10, "params": { "model": "gpt-6-sol" } }

112```112```

113 113 

114Responses echo the `id` with either `result` or `error`:114Responses echo the `id` with either `result` or `error`:

188 },188 },

189});189});

190send({ method: "initialized", params: {} });190send({ method: "initialized", params: {} });

191send({ method: "thread/start", id: 1, params: { model: "gpt-5.6-terra" } });191send({ method: "thread/start", id: 1, params: { model: "gpt-6-sol" } });

192```192```

193 193 

194## Core primitives194## Core primitives

390 390 

391Call `model/list` to discover available models and their capabilities before rendering model or personality selectors.391Call `model/list` to discover available models and their capabilities before rendering model or personality selectors.

392 392 

393The response below illustrates the structure. Available models, reasoning

394efforts, and defaults depend on the client and account; use the returned

395values rather than hard-coding this example.

396 

393```json397```json

394{ "method": "model/list", "id": 6, "params": { "limit": 20, "includeHidden": false } }398{ "method": "model/list", "id": 6, "params": { "limit": 20, "includeHidden": false } }

395{ "id": 6, "result": {399{ "id": 6, "result": {

396 "data": [{400 "data": [{

397 "id": "gpt-5.6-sol",401 "id": "gpt-6-sol",

398 "model": "gpt-5.6-sol",402 "model": "gpt-6-sol",

399 "displayName": "GPT-5.6-Sol",403 "displayName": "GPT-6 Sol",

400 "hidden": false,404 "hidden": false,

401 "defaultReasoningEffort": "low",405 "defaultReasoningEffort": "medium",

402 "supportedReasoningEfforts": [{406 "supportedReasoningEfforts": [{

403 "reasoningEffort": "low",407 "reasoningEffort": "medium",

404 "description": "Fast responses with lighter reasoning"408 "description": "Balances speed and reasoning depth for everyday tasks"

405 }],409 }],

406 "inputModalities": ["text", "image"],410 "inputModalities": ["text", "image"],

407 "supportsPersonality": true,411 "supportsPersonality": true,

489 493 

490```json494```json

491{ "method": "thread/start", "id": 10, "params": {495{ "method": "thread/start", "id": 10, "params": {

492 "model": "gpt-5.6-terra",496 "model": "gpt-6-sol",

493 "cwd": "/Users/me/project",497 "cwd": "/Users/me/project",

494 "approvalPolicy": "never",498 "approvalPolicy": "never",

495 "sandbox": "workspaceWrite",499 "sandbox": "workspaceWrite",

974 "writableRoots": ["/Users/me/project"],978 "writableRoots": ["/Users/me/project"],

975 "networkAccess": true979 "networkAccess": true

976 },980 },

977 "model": "gpt-5.6-terra",981 "model": "gpt-6-sol",

978 "effort": "medium",982 "effort": "medium",

979 "summary": "concise",983 "summary": "concise",

980 "personality": "friendly",984 "personality": "friendly",

11 11 

12GPT-5.5 retires from ChatGPT, ChatGPT Work, and Codex on all plans on12GPT-5.5 retires from ChatGPT, ChatGPT Work, and Codex on all plans on

13October 14, 2026. Review scheduled tasks that use GPT-5.5 and choose an13October 14, 2026. Review scheduled tasks that use GPT-5.5 and choose an

14available replacement before that date. For Codex with ChatGPT sign-in,14available replacement before that date. For Codex with ChatGPT sign-in, choose

15replace `gpt-5.5` with `gpt-5.6-sol` (GPT-5.6 Sol). The OpenAI API isn't15`gpt-6-sol` (GPT-6 Sol) if your plan and workspace provide access. The OpenAI API isn't

16affected. See [GPT-5.5 retirement](https://learn.chatgpt.com/docs/models#gpt-55-retirement).16affected. See [GPT-5.5 retirement](https://learn.chatgpt.com/docs/models#gpt-55-retirement) for

17plan-specific replacements.

17 18 

18</ContentModeSwitch>19</ContentModeSwitch>

19 20 

149You can also leave the model and reasoning effort on their default settings, or150You can also leave the model and reasoning effort on their default settings, or

150choose them explicitly if you want more control over how the scheduled task runs.151choose them explicitly if you want more control over how the scheduled task runs.

151 152 

152If a scheduled task uses `gpt-5.4` or `gpt-5.4-mini` with ChatGPT sign-in,153If a scheduled task uses a retired model, update it to an available replacement.

153update it before those models retire on August 31, 2026. Replace `gpt-5.4` with154See [model migration guidance](https://learn.chatgpt.com/docs/models#deprecated-codex-models).

154`gpt-5.6-terra` and `gpt-5.4-mini` with `gpt-5.6-luna`.

155 155 

156 156 

157 157 

158> Illustration: ChatGPT composer ready to create a scheduled task with 5.6 Sol Extended selected.158> Illustration: ChatGPT composer ready to create a scheduled task with 6 Sol Medium selected.

159 159 

160 160 

161 161 

11 11 

12GPT-5.5 retires from ChatGPT, ChatGPT Work, and Codex on all plans on12GPT-5.5 retires from ChatGPT, ChatGPT Work, and Codex on all plans on

13October 14, 2026. Review scheduled tasks that use GPT-5.5 and choose an13October 14, 2026. Review scheduled tasks that use GPT-5.5 and choose an

14available replacement before that date. For Codex with ChatGPT sign-in,14available replacement before that date. For Codex with ChatGPT sign-in, choose

15replace `gpt-5.5` with `gpt-5.6-sol` (GPT-5.6 Sol). The OpenAI API isn't15`gpt-6-sol` (GPT-6 Sol) if your plan and workspace provide access. The OpenAI API isn't

16affected. See [GPT-5.5 retirement](https://learn.chatgpt.com/docs/models#gpt-55-retirement).16affected. See [GPT-5.5 retirement](https://learn.chatgpt.com/docs/models#gpt-55-retirement) for

17plan-specific replacements.

17 18 

18</ContentModeSwitch>19</ContentModeSwitch>

19 20 

149You can also leave the model and reasoning effort on their default settings, or150You can also leave the model and reasoning effort on their default settings, or

150choose them explicitly if you want more control over how the scheduled task runs.151choose them explicitly if you want more control over how the scheduled task runs.

151 152 

152If a scheduled task uses `gpt-5.4` or `gpt-5.4-mini` with ChatGPT sign-in,153If a scheduled task uses a retired model, update it to an available replacement.

153update it before those models retire on August 31, 2026. Replace `gpt-5.4` with154See [model migration guidance](https://learn.chatgpt.com/docs/models#deprecated-codex-models).

154`gpt-5.6-terra` and `gpt-5.4-mini` with `gpt-5.6-luna`.

155 155 

156 156 

157 157 

158> Illustration: ChatGPT composer ready to create a scheduled task with 5.6 Sol Extended selected.158> Illustration: ChatGPT composer ready to create a scheduled task with 6 Sol Medium selected.

159 159 

160 160 

161 161 

95 95 

96with Codex() as codex:96with Codex() as codex:

97 thread = codex.thread_start(97 thread = codex.thread_start(

98 model="gpt-5.6-terra",98 model="gpt-6-sol",

99 sandbox=Sandbox.workspace_write,99 sandbox=Sandbox.workspace_write,

100 )100 )

101 result = thread.run("Make a plan to diagnose and fix the CI failures")101 result = thread.run("Make a plan to diagnose and fix the CI failures")

112 112 

113async def main() -> None:113async def main() -> None:

114 async with AsyncCodex() as codex:114 async with AsyncCodex() as codex:

115 thread = await codex.thread_start(model="gpt-5.6-terra")115 thread = await codex.thread_start(model="gpt-6-sol")

116 result = await thread.run("Implement the plan")116 result = await thread.run("Implement the plan")

117 print(result.final_response)117 print(result.final_response)

118 118 

162request a specific model or reasoning effort in your prompt,162request a specific model or reasoning effort in your prompt,

163configure `[agents]` defaults in `config.toml`, or set `model` and163configure `[agents]` defaults in `config.toml`, or set `model` and

164`model_reasoning_effort` directly in the custom agent file.164`model_reasoning_effort` directly in the custom agent file.

165For example, use `gpt-5.6-terra` for fast scans or a higher-effort `gpt-5.6` configuration for more demanding reasoning.165For example, use `gpt-6-luna` for fast scans or a higher-effort `gpt-6-sol` configuration for more demanding reasoning.

166 166 

167For most tasks in Codex, start with 167For most tasks in Codex, start with

168 `gpt-5.6`. Use 168 `gpt-6-sol`. Use

169 `gpt-5.6-terra` when you want169 `gpt-6-luna` when you want a faster, lower-cost option for lighter subagent

170 a faster, lower-cost option for lighter subagent work.170 work.

171 171 

172### Model choice172### Model choice

173 173 

174- **`gpt-5.6`**: Start here for demanding agents. It's strongest for ambiguous, multi-step work that needs planning, tool use, validation, and follow-through across a larger context.174- **`gpt-6-sol`**: Start here for demanding agents. It's strongest for ambiguous, multi-step work that needs planning, tool use, validation, and follow-through across a larger context.

175- **`gpt-5.6-terra`**: Use for agents that favor speed and efficiency over depth, such as exploration, read-heavy scans, large-file review, or processing supporting documents. It works well for parallel workers that return distilled results to the main agent.175- **`gpt-6-luna`**: Use for fast, narrowly scoped agents handling clear, repeatable, or high-volume work.

176- **`gpt-5.6-luna`**: Use for fast, narrowly scoped agents handling clear, repeatable, or high-volume work.

177 176 

178### Reasoning effort (`model_reasoning_effort`)177### Reasoning effort (`model_reasoning_effort`)

179 178 

179For explicit model settings, start with `medium` for GPT-6 Sol, `high` for

180GPT-6 Luna, or `low` for GPT-6 Astra. Adjust for the task using a level the

181selected model supports.

182 

180- **`ultra`**: Use for the deepest reasoning when the selected model supports183- **`ultra`**: Use for the deepest reasoning when the selected model supports

181 it.184 it.

182- **`max`** and **`xhigh`**: Use for especially demanding reasoning when the185- **`max`** and **`xhigh`**: Use for especially demanding reasoning when the

183 selected model supports these levels.186 selected model supports these levels.

184- **`high`**: Use when an agent needs to trace complex logic, check assumptions, or work through edge cases (for example, reviewer or security-focused agents).187- **`high`**: Use when an agent needs to trace complex logic, check assumptions, or work through edge cases (for example, reviewer or security-focused agents).

185- **`medium`**: A balanced default for most agents.188- **`medium`**: Balances speed and depth; the starting point for GPT-6 Sol.

186- **`low`**: Use when the task is straightforward and speed matters most.189- **`low`**: Use when the task is straightforward and speed matters most.

187 190 

188Higher reasoning effort increases response time and token usage, but it can improve quality for complex work. For details, see [Models](https://learn.chatgpt.com/docs/models), [Config basics](https://learn.chatgpt.com/docs/config-file/config-basic), and [Configuration Reference](https://learn.chatgpt.com/docs/config-file/config-reference).191Higher reasoning effort increases response time and token usage, but it can improve quality for complex work. For details, see [Models](https://learn.chatgpt.com/docs/models), [Config basics](https://learn.chatgpt.com/docs/config-file/config-basic), and [Configuration Reference](https://learn.chatgpt.com/docs/config-file/config-reference).

420```toml423```toml

421name = "pr_explorer"424name = "pr_explorer"

422description = "Read-only codebase explorer for gathering evidence before changes are proposed."425description = "Read-only codebase explorer for gathering evidence before changes are proposed."

423model = "gpt-5.6-luna"426model = "gpt-6-luna"

424model_reasoning_effort = "medium"427model_reasoning_effort = "high"

425sandbox_mode = "read-only"428sandbox_mode = "read-only"

426developer_instructions = """429developer_instructions = """

427Stay in exploration mode.430Stay in exploration mode.

435```toml438```toml

436name = "reviewer"439name = "reviewer"

437description = "PR reviewer focused on correctness, security, and missing tests."440description = "PR reviewer focused on correctness, security, and missing tests."

438model = "gpt-5.6-terra"441model = "gpt-6-sol"

439model_reasoning_effort = "high"442model_reasoning_effort = "medium"

440sandbox_mode = "read-only"443sandbox_mode = "read-only"

441developer_instructions = """444developer_instructions = """

442Review code like an owner.445Review code like an owner.

450```toml453```toml

451name = "docs_researcher"454name = "docs_researcher"

452description = "Documentation specialist that uses the docs MCP server to verify APIs and framework behavior."455description = "Documentation specialist that uses the docs MCP server to verify APIs and framework behavior."

453model = "gpt-5.6-luna"456model = "gpt-6-luna"

454model_reasoning_effort = "medium"457model_reasoning_effort = "high"

455sandbox_mode = "read-only"458sandbox_mode = "read-only"

456developer_instructions = """459developer_instructions = """

457Use the docs MCP server to confirm APIs, options, and version-specific behavior.460Use the docs MCP server to confirm APIs, options, and version-specific behavior.

485```toml488```toml

486name = "code_mapper"489name = "code_mapper"

487description = "Read-only codebase explorer for locating the relevant frontend and backend code paths."490description = "Read-only codebase explorer for locating the relevant frontend and backend code paths."

488model = "gpt-5.6-luna"491model = "gpt-6-luna"

489model_reasoning_effort = "medium"492model_reasoning_effort = "high"

490sandbox_mode = "read-only"493sandbox_mode = "read-only"

491developer_instructions = """494developer_instructions = """

492Map the code that owns the failing UI flow.495Map the code that owns the failing UI flow.

499```toml502```toml

500name = "browser_debugger"503name = "browser_debugger"

501description = "UI debugger that uses browser tooling to reproduce issues and capture evidence."504description = "UI debugger that uses browser tooling to reproduce issues and capture evidence."

502model = "gpt-5.6-terra"505model = "gpt-6-sol"

503model_reasoning_effort = "high"506model_reasoning_effort = "medium"

504sandbox_mode = "workspace-write"507sandbox_mode = "workspace-write"

505developer_instructions = """508developer_instructions = """

506Reproduce the issue in the browser, capture exact steps, and report what the UI actually does.509Reproduce the issue in the browser, capture exact steps, and report what the UI actually does.

518```toml521```toml

519name = "ui_fixer"522name = "ui_fixer"

520description = "Implementation-focused agent for small, targeted fixes after the issue is understood."523description = "Implementation-focused agent for small, targeted fixes after the issue is understood."

521model = "gpt-5.6-luna"524model = "gpt-6-luna"

522model_reasoning_effort = "medium"525model_reasoning_effort = "high"

523developer_instructions = """526developer_instructions = """

524Own the fix once the issue is reproduced.527Own the fix once the issue is reproduced.

525Make the smallest defensible change, keep unrelated files untouched, and validate only the behavior you changed.528Make the smallest defensible change, keep unrelated files untouched, and validate only the behavior you changed.

18 18 

19```toml19```toml

20# ~/.codex/deep-review.config.toml20# ~/.codex/deep-review.config.toml

21model = "gpt-5.6-sol"21model = "gpt-6-sol"

22model_reasoning_effort = "xhigh"22model_reasoning_effort = "medium"

23approval_policy = "on-request"23approval_policy = "on-request"

24model_catalog_json = "/Users/me/.codex/model-catalogs/deep-review.json"24model_catalog_json = "/Users/me/.codex/model-catalogs/deep-review.json"

25```25```

52 52 

53```shell53```shell

54# Dedicated flag54# Dedicated flag

55codex --model gpt-5.6-terra55codex --model gpt-6-sol

56 56 

57# Generic key/value override (value is TOML, not JSON)57# Generic key/value override (value is TOML, not JSON)

58codex --config model='"gpt-5.6-terra"'58codex --config model='"gpt-6-sol"'

59codex --config sandbox_workspace_write.network_access=true59codex --config sandbox_workspace_write.network_access=true

60codex --config 'shell_environment_policy.include_only=["PATH","HOME"]'60codex --config 'shell_environment_policy.include_only=["PATH","HOME"]'

61```61```

164Define additional providers and point `model_provider` at them:164Define additional providers and point `model_provider` at them:

165 165 

166```toml166```toml

167model = "gpt-5.6-terra"167model = "gpt-6-sol"

168model_provider = "proxy"168model_provider = "proxy"

169 169 

170[model_providers.proxy]170[model_providers.proxy]

53Choose the model Codex uses by default in the CLI and IDE.53Choose the model Codex uses by default in the CLI and IDE.

54 54 

55```toml55```toml

56model = "gpt-5.6"56model = "gpt-6-sol"

57```57```

58 58 

59 59 

115Tune how much reasoning effort the model applies when supported.115Tune how much reasoning effort the model applies when supported.

116 116 

117```toml117```toml

118model_reasoning_effort = "high"118model_reasoning_effort = "medium"

119```119```

120 120 

121#### Communication style121#### Communication style

18 18 

19```toml19```toml

20# ~/.codex/deep-review.config.toml20# ~/.codex/deep-review.config.toml

21model = "gpt-5.6-sol"21model = "gpt-6-sol"

22model_reasoning_effort = "xhigh"22model_reasoning_effort = "medium"

23approval_policy = "on-request"23approval_policy = "on-request"

24model_catalog_json = "/Users/me/.codex/model-catalogs/deep-review.json"24model_catalog_json = "/Users/me/.codex/model-catalogs/deep-review.json"

25```25```

52 52 

53```shell53```shell

54# Dedicated flag54# Dedicated flag

55codex --model gpt-5.6-terra55codex --model gpt-6-sol

56 56 

57# Generic key/value override (value is TOML, not JSON)57# Generic key/value override (value is TOML, not JSON)

58codex --config model='"gpt-5.6-terra"'58codex --config model='"gpt-6-sol"'

59codex --config sandbox_workspace_write.network_access=true59codex --config sandbox_workspace_write.network_access=true

60codex --config 'shell_environment_policy.include_only=["PATH","HOME"]'60codex --config 'shell_environment_policy.include_only=["PATH","HOME"]'

61```61```

164Define additional providers and point `model_provider` at them:164Define additional providers and point `model_provider` at them:

165 165 

166```toml166```toml

167model = "gpt-5.6-terra"167model = "gpt-6-sol"

168model_provider = "proxy"168model_provider = "proxy"

169 169 

170[model_providers.proxy]170[model_providers.proxy]

53Choose the model Codex uses by default in the CLI and IDE.53Choose the model Codex uses by default in the CLI and IDE.

54 54 

55```toml55```toml

56model = "gpt-5.6"56model = "gpt-6-sol"

57```57```

58 58 

59 59 

115Tune how much reasoning effort the model applies when supported.115Tune how much reasoning effort the model applies when supported.

116 116 

117```toml117```toml

118model_reasoning_effort = "high"118model_reasoning_effort = "medium"

119```119```

120 120 

121#### Communication style121#### Communication style

32 {32 {

33 key: "model",33 key: "model",

34 type: "string",34 type: "string",

35 description: "Model to use (e.g., `gpt-5.6-sol`).",35 description: "Model to use (e.g., `gpt-6-sol`).",

36 },36 },

37 {37 {

38 key: "review_model",38 key: "review_model",

1089 },1089 },

1090 {1090 {

1091 key: "model_reasoning_effort",1091 key: "model_reasoning_effort",

1092 type: "minimal | low | medium | high | xhigh",1092 type: "string",

1093 description:1093 description:

1094 "Adjust reasoning effort for supported models (Responses API only; `xhigh` is model-dependent).",1094 "Reasoning effort advertised by the selected model, such as `low`, `medium`, `high`, `xhigh`, `max`, or `ultra`. Available levels depend on the model and client.",

1095 },1095 },

1096 {1096 {

1097 key: "plan_mode_reasoning_effort",1097 key: "plan_mode_reasoning_effort",

1098 type: "none | minimal | low | medium | high | xhigh",1098 type: "string",

1099 description:1099 description:

1100 "Plan-mode-specific reasoning override. When unset, Plan mode uses its built-in preset default.",1100 "Plan-mode-specific reasoning override using a level supported by the selected model. When unset, Plan mode uses its built-in preset default.",

1101 },1101 },

1102 {1102 {

1103 key: "model_reasoning_summary",1103 key: "model_reasoning_summary",

31 31 

32################################################################################32################################################################################

33 33 

34# Primary model used by Codex. Recommended example for most users: "gpt-5.6".34# Primary model used by Codex. Recommended example for most users: "gpt-6-sol".

35 35 

36model = "gpt-5.6"36model = "gpt-6-sol"

37 37 

38# Communication style for supported models. Allowed values: none | friendly | pragmatic38# Communication style for supported models. Allowed values: none | friendly | pragmatic

39 39 

41 41 

42# Optional model override for /review. Default: unset (uses current session model).42# Optional model override for /review. Default: unset (uses current session model).

43 43 

44# review_model = "gpt-5.6"44# review_model = "gpt-6-sol"

45 45 

46# Provider id selected from [model_providers]. Default: "openai".46# Provider id selected from [model_providers]. Default: "openai".

47 47 

79 79 

80################################################################################80################################################################################

81 81 

82# Reasoning effort: minimal | low | medium | high | xhigh82# Reasoning effort advertised by the selected model; supported levels vary by model and client.

83 83 

84# model_reasoning_effort = "medium"84# model_reasoning_effort = "medium"

85 85 

86# Optional override used when Codex runs in plan mode: none | minimal | low | medium | high | xhigh86# Optional reasoning effort override for Plan mode; use a level supported by the selected model.

87 87 

88# plan_mode_reasoning_effort = "high"88# plan_mode_reasoning_effort = "high"

89 89 

359 359 

360# Default model for spawned agents. An explicit spawn model takes precedence.360# Default model for spawned agents. An explicit spawn model takes precedence.

361 361 

362# default_subagent_model = "gpt-5.6-terra"362# default_subagent_model = "gpt-6-sol"

363 363 

364# Default reasoning effort for spawned agents. An explicit spawn effort takes precedence.364# Default reasoning effort for spawned agents. An explicit spawn effort takes precedence.

365 365 

366# default_subagent_reasoning_effort = "high"366# default_subagent_reasoning_effort = "medium"

367 367 

368# Record a model-visible message when an agent turn is interrupted. Default: true368# Record a model-visible message when an agent turn is interrupted. Default: true

369 369 

671 671 

672# [tui.model_availability_nux]672# [tui.model_availability_nux]

673 673 

674# "gpt-5.6-terra" = 1674# "gpt-6-sol" = 1

675 675 

676# Enable or disable analytics for this machine. When unset, Codex uses its default behavior.676# Enable or disable analytics for this machine. When unset, Codex uses its default behavior.

677 677 

697 697 

698# "hide_gpt-5.1-codex-max_migration_prompt" = true698# "hide_gpt-5.1-codex-max_migration_prompt" = true

699 699 

700# model_migrations = { "gpt-5.4" = "gpt-5.6-terra" }700# model_migrations = { "gpt-5.4" = "gpt-6-sol" }

701 701 

702################################################################################702################################################################################

703 703 

1079 1079 

1080# For example, a CI profile could live at $CODEX_HOME/ci.config.toml:1080# For example, a CI profile could live at $CODEX_HOME/ci.config.toml:

1081 1081 

1082# model = "gpt-5.6-terra"1082# model = "gpt-6-sol"

1083 1083 

1084# approval_policy = "on-request"1084# approval_policy = "on-request"

1085 1085 

32 {32 {

33 key: "model",33 key: "model",

34 type: "string",34 type: "string",

35 description: "Model to use (e.g., `gpt-5.6-sol`).",35 description: "Model to use (e.g., `gpt-6-sol`).",

36 },36 },

37 {37 {

38 key: "review_model",38 key: "review_model",

1089 },1089 },

1090 {1090 {

1091 key: "model_reasoning_effort",1091 key: "model_reasoning_effort",

1092 type: "minimal | low | medium | high | xhigh",1092 type: "string",

1093 description:1093 description:

1094 "Adjust reasoning effort for supported models (Responses API only; `xhigh` is model-dependent).",1094 "Reasoning effort advertised by the selected model, such as `low`, `medium`, `high`, `xhigh`, `max`, or `ultra`. Available levels depend on the model and client.",

1095 },1095 },

1096 {1096 {

1097 key: "plan_mode_reasoning_effort",1097 key: "plan_mode_reasoning_effort",

1098 type: "none | minimal | low | medium | high | xhigh",1098 type: "string",

1099 description:1099 description:

1100 "Plan-mode-specific reasoning override. When unset, Plan mode uses its built-in preset default.",1100 "Plan-mode-specific reasoning override using a level supported by the selected model. When unset, Plan mode uses its built-in preset default.",

1101 },1101 },

1102 {1102 {

1103 key: "model_reasoning_summary",1103 key: "model_reasoning_summary",

31 31 

32################################################################################32################################################################################

33 33 

34# Primary model used by Codex. Recommended example for most users: "gpt-5.6".34# Primary model used by Codex. Recommended example for most users: "gpt-6-sol".

35 35 

36model = "gpt-5.6"36model = "gpt-6-sol"

37 37 

38# Communication style for supported models. Allowed values: none | friendly | pragmatic38# Communication style for supported models. Allowed values: none | friendly | pragmatic

39 39 

41 41 

42# Optional model override for /review. Default: unset (uses current session model).42# Optional model override for /review. Default: unset (uses current session model).

43 43 

44# review_model = "gpt-5.6"44# review_model = "gpt-6-sol"

45 45 

46# Provider id selected from [model_providers]. Default: "openai".46# Provider id selected from [model_providers]. Default: "openai".

47 47 

79 79 

80################################################################################80################################################################################

81 81 

82# Reasoning effort: minimal | low | medium | high | xhigh82# Reasoning effort advertised by the selected model; supported levels vary by model and client.

83 83 

84# model_reasoning_effort = "medium"84# model_reasoning_effort = "medium"

85 85 

86# Optional override used when Codex runs in plan mode: none | minimal | low | medium | high | xhigh86# Optional reasoning effort override for Plan mode; use a level supported by the selected model.

87 87 

88# plan_mode_reasoning_effort = "high"88# plan_mode_reasoning_effort = "high"

89 89 

359 359 

360# Default model for spawned agents. An explicit spawn model takes precedence.360# Default model for spawned agents. An explicit spawn model takes precedence.

361 361 

362# default_subagent_model = "gpt-5.6-terra"362# default_subagent_model = "gpt-6-sol"

363 363 

364# Default reasoning effort for spawned agents. An explicit spawn effort takes precedence.364# Default reasoning effort for spawned agents. An explicit spawn effort takes precedence.

365 365 

366# default_subagent_reasoning_effort = "high"366# default_subagent_reasoning_effort = "medium"

367 367 

368# Record a model-visible message when an agent turn is interrupted. Default: true368# Record a model-visible message when an agent turn is interrupted. Default: true

369 369 

671 671 

672# [tui.model_availability_nux]672# [tui.model_availability_nux]

673 673 

674# "gpt-5.6-terra" = 1674# "gpt-6-sol" = 1

675 675 

676# Enable or disable analytics for this machine. When unset, Codex uses its default behavior.676# Enable or disable analytics for this machine. When unset, Codex uses its default behavior.

677 677 

697 697 

698# "hide_gpt-5.1-codex-max_migration_prompt" = true698# "hide_gpt-5.1-codex-max_migration_prompt" = true

699 699 

700# model_migrations = { "gpt-5.4" = "gpt-5.6-terra" }700# model_migrations = { "gpt-5.4" = "gpt-6-sol" }

701 701 

702################################################################################702################################################################################

703 703 

1079 1079 

1080# For example, a CI profile could live at $CODEX_HOME/ci.config.toml:1080# For example, a CI profile could live at $CODEX_HOME/ci.config.toml:

1081 1081 

1082# model = "gpt-5.6-terra"1082# model = "gpt-6-sol"

1083 1083 

1084# approval_policy = "on-request"1084# approval_policy = "on-request"

1085 1085 

768defaults apply again the next time the client starts.768defaults apply again the next time the client starts.

769 769 

770If a managed default, macOS MDM profile, or saved configuration pins770If a managed default, macOS MDM profile, or saved configuration pins

771`gpt-5.5` for Codex users signed in with ChatGPT, replace it with771`gpt-5.5` for Codex users signed in with ChatGPT, replace it with an available

772`gpt-5.6-sol` before October 14, 2026. GPT-5.5 retires from ChatGPT,772model before October 14, 2026. Choose `gpt-6-sol` once an administrator has

773enabled it for the affected users. GPT-5.5 retires from ChatGPT,

773ChatGPT Work, and Codex on all plans on that date. The OpenAI API isn't774ChatGPT Work, and Codex on all plans on that date. The OpenAI API isn't

774affected. See [workspace model availability](https://learn.chatgpt.com/docs/enterprise/workspace-model-availability#prepare-for-the-gpt-55-retirement).775affected. See [workspace model availability](https://learn.chatgpt.com/docs/enterprise/workspace-model-availability#prepare-for-the-gpt-55-retirement).

775 776 

776If a managed default, macOS MDM profile, or saved configuration pins `gpt-5.4`777For configurations that still pin `gpt-5.4` or `gpt-5.4-mini`, follow the

777or `gpt-5.4-mini` for users signed in with ChatGPT, update it before August 31, 2026. Replace `gpt-5.4` with `gpt-5.6-terra` and `gpt-5.4-mini` with778[GPT-5.4 migration guidance](https://learn.chatgpt.com/docs/enterprise/workspace-model-availability#prepare-for-the-gpt-54-retirement).

778`gpt-5.6-luna`. The OpenAI API and Codex authenticated with your own API key

779aren't affected. See [workspace model

780availability](https://learn.chatgpt.com/docs/enterprise/workspace-model-availability#prepare-for-the-gpt-54-retirement).

781 779 

782Make sure your managed defaults meet your requirements; the local runtime780Make sure your managed defaults meet your requirements; the local runtime

783rejects disallowed values.781rejects disallowed values.

44This setting can pin Fast mode on or off for managed local Codex clients; it44This setting can pin Fast mode on or off for managed local Codex clients; it

45isn't a starting default and can't override workspace or product availability.45isn't a starting default and can't override workspace or product availability.

46 46 

47## GPT-6 Sol and Luna in Enterprise

48 

49GPT-6 Sol and GPT-6 Luna are off by default in Enterprise workspaces at

50launch. An administrator must enable each model before members can select

51it. Review your [workspace model settings](https://help.openai.com/en/articles/8411955)

52and confirm access on each client. Choosing a model in local configuration

53doesn't override workspace controls.

54 

47## GPT-6 Astra in Enterprise55## GPT-6 Astra in Enterprise

48 56 

49During the initial rollout, your organization must have Daybreak access before57During the initial rollout, your organization must have Daybreak access before

73retirement does not apply to the OpenAI API.81retirement does not apply to the OpenAI API.

74 82 

75Before October 14, review workspace defaults for ChatGPT, ChatGPT Work, and83Before October 14, review workspace defaults for ChatGPT, ChatGPT Work, and

76Codex and choose an available replacement. For Codex with ChatGPT sign-in,84Codex and choose an available replacement for each surface. For Work and Codex

77replace `gpt-5.5` with `gpt-5.6-sol` (GPT-5.6 Sol) in workspace defaults,85with ChatGPT sign-in, choose `gpt-6-sol` (GPT-6 Sol) once an administrator has

86enabled it for the affected users. Replace `gpt-5.5` in workspace defaults,

78saved model settings, managed configurations, custom agents, and scheduled87saved model settings, managed configurations, custom agents, and scheduled

79tasks. Check scripts and commands that explicitly select `gpt-5.5` too.88tasks. Check scripts and commands that explicitly select `gpt-5.5` too.

80 89 

86 95 

87## Prepare for the GPT-5.4 retirement96## Prepare for the GPT-5.4 retirement

88 97 

89On August 31, 2026, GPT-5.4 and GPT-5.4 mini retire from Codex for users signed98GPT-5.4 and GPT-5.4 mini retired from Codex for users signed in with ChatGPT

90in with ChatGPT. Update affected workspace defaults, saved model settings,99on August 31, 2026. Update any remaining workspace defaults, saved model

91managed configurations, custom agents, and scheduled tasks before then:100settings, managed configurations, custom agents, and scheduled tasks with

101models available to the affected users' plans and clients:

92 102 

93- Replace `gpt-5.4` with `gpt-5.6-terra` (GPT-5.6 Terra).103- Replace `gpt-5.4` with `gpt-6-sol` (GPT-6 Sol) when available.

94- Replace `gpt-5.4-mini` with `gpt-5.6-luna` (GPT-5.6 Luna).104- Replace `gpt-5.4-mini` with `gpt-6-luna` (GPT-6 Luna) when available. In

105 Enterprise and Edu, an administrator must enable Luna first.

95 106 

96The OpenAI API and Codex authenticated with your own API key aren't affected.107The OpenAI API and Codex authenticated with your own API key aren't affected.

97See [Codex models](https://learn.chatgpt.com/docs/models#deprecated-codex-models) and108See [Codex models](https://learn.chatgpt.com/docs/models#deprecated-codex-models) and

168 168 

169Choose [GPT-6 Astra](https://learn.chatgpt.com/docs/models#gpt-6-astra) for demanding work that needs169Choose [GPT-6 Astra](https://learn.chatgpt.com/docs/models#gpt-6-astra) for demanding work that needs

170careful reasoning, visual judgment, or a polished final file. For simpler tasks,170careful reasoning, visual judgment, or a polished final file. For simpler tasks,

171consider Sol, Terra, or Luna. Select from the models171consider Sol or Luna. Select from the models

172available in your model selector and check [plan usage](https://learn.chatgpt.com/docs/pricing)172available in your model selector and check [plan usage](https://learn.chatgpt.com/docs/pricing)

173before starting a large task.173before starting a large task.

174 174 

1# Model selection

2 

3> For the complete documentation index, see [llms.txt](https://learn.chatgpt.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to the page URL.

4 

5## Meet the models

6 

7<ModelOverview />

8 

9Availability, tools, reasoning settings, and usage limits differ by product and

10model version. Check the [models available in ChatGPT](https://learn.chatgpt.com/docs/models) or the

11[API model catalog](https://developers.openai.com/api/docs/models).

12 

13## Find the right model for your workflow

14 

15Choose your work and task and get a recommendation.

16 

17<ModelRecommender client:load />

18 

19## How to think about models and reasoning effort

20 

21 

22 

23 

24 

25 Luna is our most efficient model, while Astra is our state-of-the-art, most

26 powerful model. If you don't need to think about usage or how long it takes to

27 complete a task, you can default to Astra. To optimize usage, use the guidance

28 below to choose a model and reasoning effort for your needs.

29 

30 

31 

321. **Luna · Low**

33 

34 Fine-grained edits, well-scoped problem-solving, and simple data extraction.

35 

362. **Luna · Medium**

37 

38 Creating from clear briefs and making coordinated updates to existing work.

39 

403. **Luna · Extra high**

41 

42 Finding current context across multiple apps, prioritizing work, and solving problems with clear constraints.

43 

444. **Sol · Low**

45 

46 Focused writing and editing, fact-checking, and straightforward work in apps.

47 

485. **Sol · Medium**

49 

50 Everyday coding, research, and workflows that need judgment and completeness.

51 

526. **Sol · Extra high**

53 

54 Deeper analysis, thorough verification, and careful review of documents, data, and code.

55 

567. **Astra · Low**

57 

58 Concise writing and content adaptation that preserve facts and nuance.

59 

608. **Astra · Medium**

61 

62 Ambitious projects that need broad context, reliable interactions, and complete results.

63 

649. **Astra · Extra high**

65 

66 Demanding analysis and complex deliverables with exacting requirements.

67 

68### Experiment

69 

70Treat the guidance on this page as a starting point. The best way to find the right

71model for your workflow is to experiment with different models and reasoning

72settings to see what works.

73 

74Start by considering:

75 

76- **How often does your workflow run?** A frequent automation makes usage and cost

77 add up faster than an occasional project.

78- **How quickly do you need the result?** A task you're waiting on may need a

79 faster setting than one that runs overnight.

80- **How will you use the output?** A draft for your review may need less polish

81 than something you'll share externally.

82- **How important is the quality of the result?** Depending on your use case or

83 industry, you might want to use a stronger model to put an emphasis on quality.

84 

85If you can, experiment using the same inputs to compare results and keep the

86lightest setting that meets your quality bar.

8on all plans, including consumer, Business, Enterprise, and Edu plans. This8on all plans, including consumer, Business, Enterprise, and Edu plans. This

9retirement does not apply to the OpenAI API.9retirement does not apply to the OpenAI API.

10 10 

11If you use Codex with ChatGPT sign-in, switch to **GPT-5.6 Sol**11If you use Codex with ChatGPT sign-in, choose an available replacement before

12(`gpt-5.6-sol`) before October 14. Replace `gpt-5.5` in workspace defaults,12October 14:

13saved model settings, managed configurations, custom agents, scheduled tasks,13 

14and scripts that select a model.14- On Plus, Pro, Business, Enterprise, and Edu plans, choose **GPT-6 Sol**

15 (`gpt-6-sol`) when available.

16- On Free and Go plans, choose **GPT-6 Luna** (`gpt-6-luna`) in the desktop app

17 when available.

18 

19Replace `gpt-5.5` in workspace defaults, saved model settings, managed

20configurations, custom agents, scheduled tasks, and scripts that select a model.

15 21 

16See [workspace model availability](https://learn.chatgpt.com/docs/enterprise/workspace-model-availability#prepare-for-the-gpt-55-retirement)22See [workspace model availability](https://learn.chatgpt.com/docs/enterprise/workspace-model-availability#prepare-for-the-gpt-55-retirement)

17for administrator guidance.23for administrator guidance.

25 31 

26## Choose a model32## Choose a model

27 33 

28In the ChatGPT desktop app, use the model and reasoning control beneath the34In Work or Codex in the ChatGPT desktop app, use the model and reasoning control

29composer to choose an available model and adjust its reasoning effort.35beneath the composer to choose an available model and adjust its reasoning effort.

30 36 

31Higher reasoning effort can improve results for complex tasks, but it takes37Higher reasoning effort can improve results for complex tasks, but it takes

32longer and uses more tokens. Start with the default effort and increase it when38longer and uses more tokens. Start with the default effort and increase it when

89`--model` or its `-m` alias:95`--model` or its `-m` alias:

90 96 

91```bash97```bash

92codex --model gpt-5.698codex --model gpt-6-sol

93```99```

94 100 

95 101 

96The same option works with non-interactive runs. For example:102The same option works with non-interactive runs. For example:

97 103 

98```bash104```bash

99codex exec -m gpt-5.6 "Review the current changes"105codex exec -m gpt-6-sol "Review the current changes"

100```106```

101 107 

102 108 

160 166 

161## Recommended models167## Recommended models

162 168 

169GPT-6 Sol and GPT-6 Luna bring improved coding, factual reliability, and

170communication to lower-cost models. Use Sol for complex coding and agentic

171workflows, and Luna for focused, repeatable tasks. Select `gpt-6-sol` or

172`gpt-6-luna` in your model picker or saved configuration when available.

173 

174In ChatGPT, GPT-6 Sol and GPT-6 Luna are available in Work and Codex. They

175aren't available in Chat.

176 

177GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna remain available during the rollout. Selecting a

178new model doesn't change workspace permissions or grant access to it.

179 

163<a id="app-compare-models"></a>180<a id="app-compare-models"></a>

164 181 

165 182 

202 219 

203<ModelDetails220<ModelDetails

204 client:load221 client:load

205 name="gpt-5.6-sol"222 name="gpt-6-sol"

206 slug="gpt-5.6-sol"223 slug="gpt-6-sol"

207 imageLabel="5.6 Sol"224 imageLabel="6 Sol"

208 wallpaperUrl="/images/api/models/gpt-5.6-sol.webp"225 wallpaperUrl="/images/api/models/gpt-5.6-sol.webp"

209 description="The most capable GPT-5.6 model for complex coding, computer use, research, and cybersecurity."226 description="Built for complex coding and agentic workflows, with stronger factual reliability and clearer communication than GPT-5.6 Sol."

210 data={{227 data={{

211 features: [228 features: [

212 {229 {

225 value: "",242 value: "",

226 icons: ["openai.Flash", "openai.Flash"],243 icons: ["openai.Flash", "openai.Flash"],

227 },244 },

228 { title: "ChatGPT desktop app", value: true },245 { title: "Work and Codex in the desktop app", value: true },

229 { title: "ChatGPT web", value: true },246 { title: "ChatGPT Work on the web", value: true },

230 { title: "Codex CLI", value: true },247 { title: "Codex CLI", value: true },

231 { title: "Codex IDE extension", value: true },248 { title: "Codex IDE extension", value: true },

232 { title: "Codex cloud", value: true },

233 { title: "ChatGPT Credits", value: true },249 { title: "ChatGPT Credits", value: true },

234 { title: "API Access", value: true },250 { title: "API Access", value: true },

235 ],251 ],

238 254 

239<ModelDetails255<ModelDetails

240 client:load256 client:load

241 name="gpt-5.6-terra"257 name="gpt-6-luna"

242 slug="gpt-5.6-terra"258 slug="gpt-6-luna"

243 imageLabel="5.6 Terra"259 imageLabel="6 Luna"

244 wallpaperUrl="/images/api/models/gpt-5.6-terra.webp"

245 description="Balanced GPT-5.6 model for everyday work, with performance competitive with GPT-5.5 at a lower cost."

246 data={{

247 features: [

248 {

249 title: "Capability",

250 value: "",

251 icons: [

252 "openai.SparklesFilled",

253 "openai.SparklesFilled",

254 "openai.SparklesFilled",

255 "openai.SparklesFilled",

256 ],

257 },

258 {

259 title: "Speed",

260 value: "",

261 icons: ["openai.Flash", "openai.Flash", "openai.Flash"],

262 },

263 { title: "ChatGPT desktop app", value: true },

264 { title: "ChatGPT web", value: true },

265 { title: "Codex CLI", value: true },

266 { title: "Codex IDE extension", value: true },

267 { title: "Codex cloud", value: false },

268 { title: "ChatGPT Credits", value: true },

269 { title: "API Access", value: true },

270 ],

271 }}

272/>

273 

274<ModelDetails

275 client:load

276 name="gpt-5.6-luna"

277 slug="gpt-5.6-luna"

278 imageLabel="5.6 Luna"

279 wallpaperUrl="/images/api/models/gpt-5.6-luna.webp"260 wallpaperUrl="/images/api/models/gpt-5.6-luna.webp"

280 description="Fast and affordable GPT-5.6 model that delivers strong capability at the lowest cost in the family."261 description="Our most efficient model for focused, high-volume tasks, including summarization, extraction, and focused coding."

281 data={{262 data={{

282 features: [263 features: [

283 {264 {

294 value: "",275 value: "",

295 icons: ["openai.Flash", "openai.Flash", "openai.Flash", "openai.Flash"],276 icons: ["openai.Flash", "openai.Flash", "openai.Flash", "openai.Flash"],

296 },277 },

297 { title: "ChatGPT desktop app", value: true },278 { title: "Work and Codex in the desktop app", value: true },

298 { title: "ChatGPT web", value: true },279 { title: "ChatGPT Work on the web", value: true },

299 { title: "Codex CLI", value: true },280 { title: "Codex CLI", value: true },

300 { title: "Codex IDE extension", value: true },281 { title: "Codex IDE extension", value: true },

301 { title: "Codex cloud", value: false },282 { title: "Codex cloud", value: false },

310 291 

311Availability depends on the rollout, your sign-in method, and your client.292Availability depends on the rollout, your sign-in method, and your client.

312See [pricing](https://learn.chatgpt.com/docs/pricing) for plan access and usage, and293See [pricing](https://learn.chatgpt.com/docs/pricing) for plan access and usage, and

313[workspace model availability](https://learn.chatgpt.com/docs/enterprise/workspace-model-availability#gpt-6-astra-in-enterprise)294[workspace model availability](https://learn.chatgpt.com/docs/enterprise/workspace-model-availability)

314for Enterprise access.295for Enterprise access.

315 296 

316Start with the default Power setting available to your account. Move toward297Start with the default Power setting available to your account. Move toward

317 **Smarter** for deeper reasoning or **Faster** for faster, lower-cost work.298 **Smarter** for deeper reasoning or **Faster** for faster, lower-cost work.

318 Open **Advanced** when you want `gpt-5.6-luna` or a specific model, reasoning299 Open **Advanced** to choose a specific model, reasoning effort, or speed.

319 effort, or speed.

320 300 

321<ContentModeSwitch group="codex-surface" ids="app,web">301<ContentModeSwitch group="codex-surface" ids="app,web">

322 302 

323The picker illustrations show GPT-5.6 controls. For eligible Pro, Business303The six Power presets shown are Luna High, Sol Light (the starting

324($100), and Enterprise accounts, the Astra rollout updates the Power options304preset), Sol Medium, Astra Light, Astra Medium, and Astra Extra High. Some paid plans omit

325to Terra Light, Sol Light, Sol Medium, Astra Light, Astra Medium, and Astra305Astra Extra High. Advanced controls are illustrative; available options

326Extra High. Options can differ by plan and rollout stage.306and defaults vary by plan, client, workspace settings, and rollout.

327 307 

328</ContentModeSwitch>308</ContentModeSwitch>

329 309 

346 326 

347<a id="choosing-sol-terra-and-luna"></a>327<a id="choosing-sol-terra-and-luna"></a>

348 328 

349## Choosing Astra, Sol, Terra, and Luna329## Choosing Astra, Sol, and Luna

350 330 

351Choose **Astra** when a task needs the strongest capability across multiple331Choose **Astra** when a task needs the strongest capability across multiple

352steps and tools. **Sol** offers depth and polish, **Terra** suits everyday work,332steps and tools. **Sol** suits everyday work and complex coding,

353and **Luna** suits clear, repeatable tasks.333and **Luna** suits clear, repeatable tasks.

354 334 

355### Where each model shines335### Where each model shines

359 Give it the sources, templates, constraints, and checks that define a useful339 Give it the sources, templates, constraints, and checks that define a useful

360 result. Astra is better at asking focused questions and incorporating your340 result. Astra is better at asking focused questions and incorporating your

361 guidance while keeping the original goal and constraints in view.341 guidance while keeping the original goal and constraints in view.

362- **Sol, for complex, open-ended work.** Choose Sol for ambiguous, difficult, or342- **Sol, for everyday and complex work.** Choose Sol for ambiguous, difficult, or

363 high-value tasks that need extra analysis, judgment, or polish, such as343 high-value tasks that need extra analysis, judgment, or polish, such as

364 complex code changes, deep research, or polished documents. For narrower344 complex code changes, deep research, or polished documents. For narrower

365 tasks, define what done looks like to keep the work focused.345 tasks, define what done looks like to keep the work focused.

366- **Terra, the pragmatic all-rounder.** Choose Terra for everyday work that

367 needs strong reasoning and tool use when you do not need Sol's full depth. It

368 is a natural starting point for work you previously gave GPT-5.5.

369- **Luna, for clear, repeatable tasks.** Choose Luna for specific, high-volume346- **Luna, for clear, repeatable tasks.** Choose Luna for specific, high-volume

370 tasks when you know what a good result looks like, such as extraction,347 tasks when you know what a good result looks like, such as extraction,

371 classification, transformation, and structured summaries.348 classification, transformation, and structured summaries.

372 349 

373### Pick a reasoning effort350### Pick a reasoning effort

374 351 

375Use the lowest reasoning effort that produces the result you need. Increase it352Start with **Medium** for Sol, **High** for Luna, or **Light** for Astra.

376for tasks that need more planning, analysis, or checking.353In configuration, Astra's Light setting is `low`. Increase the effort for tasks

354that need more planning, analysis, or checking.

377 355 

378- **Light** in the ChatGPT desktop app, ChatGPT Work on the web, and IDE extension, or **Low** in the356- **Light** in the ChatGPT desktop app, ChatGPT Work on the web, and IDE extension, or **Low** in the

379 CLI, suits quick, well-scoped tasks.357 CLI, suits quick, well-scoped tasks.

381- **High** and **Extra High** suit difficult work with multiple steps, sources,359- **High** and **Extra High** suit difficult work with multiple steps, sources,

382 or tradeoffs.360 or tradeoffs.

383 361 

384GPT-5.5 reasoning efforts don't map exactly to GPT-5.6. Try a familiar task362Reasoning efforts don't map exactly between model generations. Try a familiar task

385at a lower setting and adjust based on the result.363at a lower setting and adjust based on the result.

386 364 

387### Know when to use Max or Ultra365### Know when to use Max or Ultra

394separate parts of a complex task in parallel. Choose it when you can divide the372separate parts of a complex task in parallel. Choose it when you can divide the

395work into meaningful parts. Most tasks do not need Max or Ultra.373work into meaningful parts. Most tasks do not need Max or Ultra.

396 374 

397If Ultra doesn't appear in the desktop app's model slider, go to375GPT-6 Luna supports reasoning efforts up to **Max**, but not **Ultra**.

398**Settings** > **Configuration**, then turn on **Ultra in model picker slider**.376 

377If your model supports Ultra but it doesn't appear in the desktop app's model

378slider, go to **Settings** > **Configuration**, then turn on

379**Ultra in model picker slider**.

399 380 

400</ContentModeSwitch>381</ContentModeSwitch>

401 382 

405 386 

406When you sign in with ChatGPT, Codex works best with the recommended models listed above.387When you sign in with ChatGPT, Codex works best with the recommended models listed above.

407 388 

408**

409 GPT-5.4 and GPT-5.4 mini retire from Codex on August 31, 2026.

410 **

411 If you sign in with ChatGPT, replace `gpt-5.4` with `gpt-5.6-terra` and

412 `gpt-5.4-mini` with `gpt-5.6-luna` in saved configurations, custom agents, and

413 scheduled tasks. The OpenAI API and Codex authenticated with your own API key

414 aren't affected.

415 

416<ToggleSection title="View other models">389<ToggleSection title="View other models">

417 390

418 391 

532## Deprecated Codex models505## Deprecated Codex models

533 506 

534GPT-5.5 retires from Codex with ChatGPT sign-in on October 14, 2026. Replace507GPT-5.5 retires from Codex with ChatGPT sign-in on October 14, 2026. Replace

535`gpt-5.5` with `gpt-5.6-sol`. See [GPT-5.5 retirement](https://learn.chatgpt.com/docs/models#gpt-55-retirement)508`gpt-5.5` with a model available to your account and client. See

536for the scope and migration checklist.509[GPT-5.5 retirement](https://learn.chatgpt.com/docs/models#gpt-55-retirement) for plan-specific

510replacements and the migration checklist.

537 511 

538The `gpt-5.4` and `gpt-5.4-mini` models retire from Codex with ChatGPT sign-in512The `gpt-5.4` and `gpt-5.4-mini` models retired from Codex with ChatGPT sign-in

539on August 31, 2026. Replace `gpt-5.4` with `gpt-5.6-terra` and513on August 31, 2026. Replace `gpt-5.4` with `gpt-6-sol` and

540`gpt-5.4-mini` with `gpt-5.6-luna` in workspace defaults, saved model514`gpt-5.4-mini` with `gpt-6-luna` when available to your plan and client.

541settings, managed configurations, custom agents, and scheduled tasks.515In Enterprise and Edu, an administrator must enable Luna first. Update workspace

516defaults, saved model settings, managed configurations, custom agents, and

517scheduled tasks with an available replacement.

542 518 

543The `gpt-5.2` and `gpt-5.3-codex` models are already deprecated in Codex when519The `gpt-5.2` and `gpt-5.3-codex` models are already deprecated in Codex when

544you sign in with ChatGPT. Update scripts, configuration files, and520you sign in with ChatGPT. Update scripts, configuration files, and

556ChatGPT desktop app, Codex CLI, or IDE extension uses a recommended model.532ChatGPT desktop app, Codex CLI, or IDE extension uses a recommended model.

557 533 

558```toml534```toml

559model = "gpt-5.6"535model = "gpt-6-sol"

560```536```

561 537 

562 538 

95 95 

96with Codex() as codex:96with Codex() as codex:

97 thread = codex.thread_start(97 thread = codex.thread_start(

98 model="gpt-5.6-terra",98 model="gpt-6-sol",

99 sandbox=Sandbox.workspace_write,99 sandbox=Sandbox.workspace_write,

100 )100 )

101 result = thread.run("Make a plan to diagnose and fix the CI failures")101 result = thread.run("Make a plan to diagnose and fix the CI failures")

112 112 

113async def main() -> None:113async def main() -> None:

114 async with AsyncCodex() as codex:114 async with AsyncCodex() as codex:

115 thread = await codex.thread_start(model="gpt-5.6-terra")115 thread = await codex.thread_start(model="gpt-6-sol")

116 result = await thread.run("Implement the plan")116 result = await thread.run("Implement the plan")

117 print(result.final_response)117 print(result.final_response)

118 118 

166 166 

167```bash167```bash

168npx @openai/codex-security scan "$REPOSITORY" \168npx @openai/codex-security scan "$REPOSITORY" \

169 --model gpt-5.6-terra \169 --model gpt-6-sol \

170 --effort high170 --effort medium

171```171```

172 172 

173Supported effort levels are `minimal`, `low`, `medium`, `high`, `xhigh`, and173Supported effort levels are `minimal`, `low`, `medium`, `high`, `xhigh`, and

152npx @openai/codex-security bulk-scan repositories.csv \152npx @openai/codex-security bulk-scan repositories.csv \

153 --output-dir /path/outside/repositories/security-scans \153 --output-dir /path/outside/repositories/security-scans \

154 --workers 4 \154 --workers 4 \

155 --model gpt-5.6-terra \155 --model gpt-6-sol \

156 --effort high156 --effort medium

157```157```

158 158 

159The same options work during interactive repository discovery:159The same options work during interactive repository discovery:

160 160 

161```bash161```bash

162npx @openai/codex-security bulk-scan --model gpt-5.6-terra --effort high162npx @openai/codex-security bulk-scan --model gpt-6-sol --effort medium

163```163```

164 164 

165Supported effort levels are `minimal`, `low`, `medium`, `high`, and `xhigh`.165Supported effort levels are `minimal`, `low`, `medium`, `high`, and `xhigh`.

419To select a different model and reasoning effort without writing TOML:419To select a different model and reasoning effort without writing TOML:

420 420 

421```bash421```bash