SpyBara
Go Premium

Documentation 2026-10-07 23:57 UTC to 2026-10-08 23:01 UTC

18 files changed +461 −164. View all changes and history on the product overview
2026
Mon 12 23:01 Fri 9 23:02 Thu 8 23:01 Wed 7 23:57 Tue 6 23:02 Mon 5 23:57 Fri 2 23:58 Thu 1 22:59
Details

426 426 

427- **macOS** uses Seatbelt policies and runs commands using `sandbox-exec` with a profile (`-p`) that corresponds to the `--sandbox` mode you selected. When restricted read access enables platform defaults, Codex appends a curated macOS platform policy (instead of broadly allowing `/System`) to preserve common tool compatibility.427- **macOS** uses Seatbelt policies and runs commands using `sandbox-exec` with a profile (`-p`) that corresponds to the `--sandbox` mode you selected. When restricted read access enables platform defaults, Codex appends a curated macOS platform policy (instead of broadly allowing `/System`) to preserve common tool compatibility.

428- **Linux** uses `bwrap` plus `seccomp` by default.428- **Linux** uses `bwrap` plus `seccomp` by default.

429- **Windows** uses the Linux sandbox implementation when running in [Windows Subsystem for Linux 2 (WSL2)](https://learn.chatgpt.com/docs/windows/wsl). WSL1 was supported through Codex `0.114`; starting in `0.115`, the Linux sandbox moved to `bwrap`, so WSL1 is no longer supported. When running natively on Windows, Codex uses a [Windows sandbox](https://learn.chatgpt.com/docs/windows/windows-sandbox#windows-sandbox) implementation.429- **Windows** supports [MXC](https://learn.chatgpt.com/docs/windows/windows-sandbox#windows-sandbox), with `elevated` and `unelevated` as legacy fallbacks. In [Windows Subsystem for Linux 2 (WSL2)](https://learn.chatgpt.com/docs/windows/wsl), Codex uses the Linux sandbox implementation. WSL1 was supported through Codex `0.114`; starting in `0.115`, the Linux sandbox moved to `bwrap`, so WSL1 is no longer supported.

430 430 

431If you use the Codex IDE extension on Windows, it supports WSL2 directly. Set the following in your VS Code settings to keep the agent inside WSL2 whenever it's available:431If you use the Codex IDE extension on Windows, it supports WSL2 directly. Set the following in your VS Code settings to keep the agent inside WSL2 whenever it's available:

432 432 


438 438 

439This ensures the IDE extension inherits Linux sandbox semantics for commands, approvals, and filesystem access even when the host OS is Windows. Learn more in the [WSL guide](https://learn.chatgpt.com/docs/windows/wsl).439This ensures the IDE extension inherits Linux sandbox semantics for commands, approvals, and filesystem access even when the host OS is Windows. Learn more in the [WSL guide](https://learn.chatgpt.com/docs/windows/wsl).

440 440 

441When running natively on Windows, configure the native sandbox mode in `config.toml`:441When running on Windows, prefer MXC when the device and policy support it by setting this in `config.toml`:

442 442 

443```toml443```toml

444[windows]444[features]

445sandbox = "unelevated" # or "elevated"445prefer_mxc = true

446# sandbox_private_desktop = true # default; set false only for compatibility

447```446```

448 447 

449See the [Windows setup guide](https://learn.chatgpt.com/docs/windows/windows-sandbox#windows-sandbox) for details.448Keep a permitted legacy implementation in `windows.sandbox` for fallback. See the [Windows setup guide](https://learn.chatgpt.com/docs/windows/windows-sandbox#prefer-mxc-with-legacy-fallback) for configuration and compatibility limits.

450 449 

451When you run Linux in a containerized environment such as Docker, the sandbox may not work if the host or container configuration blocks the namespace, setuid `bwrap`, or `seccomp` operations that Codex needs.450When you run Linux in a containerized environment such as Docker, the sandbox may not work if the host or container configuration blocks the namespace, setuid `bwrap`, or `seccomp` operations that Codex needs.

452 451 

Details

17 17 

18For supported models, Fast mode uses included subscription limits at 2.5x the18For supported models, Fast mode uses included subscription limits at 2.5x the

19Standard rate. Purchased credits and Enterprise pay-as-you-go usage are billed19Standard rate. Purchased credits and Enterprise pay-as-you-go usage are billed

20at 2x the Standard rate. These billing multipliers don't describe speed increases.20at 2x the Standard rate.

21 21 

22<a id="astra-ultrafast"></a>22<a id="astra-ultrafast"></a>

23 23 

24## Ultrafast mode24## Ultrafast mode

25 25 

26Ultrafast supports GPT-6 Astra and GPT-6.1 Sol.

27 

26GPT-6 Astra Ultrafast generates tokens up to 8x faster than GPT-6 Astra in28GPT-6 Astra Ultrafast generates tokens up to 8x faster than GPT-6 Astra in

27Standard mode in Codex. This comparison measures token generation speed,29Standard mode in Codex.

28not billing rates or overall task completion time.

29 30 

30Ultrafast is available in Codex and ChatGPT Work on Pro $500 and eligible31Ultrafast is available in Codex and ChatGPT Work on Pro $500 and eligible

31Enterprise and Edu plans. On Pro $500, Ultrafast uses your included usage first,32Enterprise and Edu plans. On Pro $500, Ultrafast uses your included usage first,

32then your available credits after that allowance runs out.33then your available credits after that allowance runs out.

33 34 

34For GPT-6 Astra, Ultrafast uses included subscription limits at 8x the35Ultrafast mode bills purchased credits and Enterprise pay-as-you-go usage at

35Standard rate. Purchased credits and Enterprise pay-as-you-go usage are billed366x the Standard rate. Ultrafast uses included subscription limits at 8x the

36at 6x the Standard rate. These billing multipliers don't describe speed increases.37Standard rate.

37Enterprise billing remains subject to the workspace's agreement.

38 38 

39For Enterprise workspaces, Ultrafast is off by default. Workspace owners can39For Enterprise workspaces, Ultrafast is off by default. Workspace owners can

40enable access for selected users or the workspace through40enable access for selected users or the workspace through


48 pricing, credits, and usage limits. See [Codex pricing](https://learn.chatgpt.com/docs/pricing) for48 pricing, credits, and usage limits. See [Codex pricing](https://learn.chatgpt.com/docs/pricing) for

49 details.49 details.

50 50 

51### Supported models

52 

53GPT-6 Astra, GPT-6 Sol, and GPT-6 Luna support Fast mode where available.

54[GPT-6.1 Sol](https://learn.chatgpt.com/docs/models#gpt-61-sol) supports Standard and Fast where

55available. Access depends on your plan, client, workspace settings, and rollout.

56See [Models](https://learn.chatgpt.com/docs/models) for model availability and51See [Models](https://learn.chatgpt.com/docs/models) for model availability and

57[Pricing](https://learn.chatgpt.com/docs/pricing#token-rates) for token rates.52[Pricing](https://learn.chatgpt.com/docs/pricing#token-rates) for token rates.

58 53 


66Other self-serve plans don't have access to Ultrafast at launch, even with61Other self-serve plans don't have access to Ultrafast at launch, even with

67purchased credits.62purchased credits.

68 63 

69Ultrafast isn't available to workspaces that require inference residency64GPT-6.1 Sol Ultrafast supports inference residency in the United States and

70outside the United States. A workspace's location alone doesn't determine65Europe (EEA + Switzerland). GPT-6 Astra Ultrafast supports inference residency

71eligibility.66in the United States only.

72 67 

73### API billing68### API billing

74 69 

Details

6what context accompanies their feedback, and which controls they use to preview6what context accompanies their feedback, and which controls they use to preview

7changes before sending an annotation to ChatGPT.7changes before sending an annotation to ChatGPT.

8 8 

9The Browser Annotation API isn't currently available in ChatGPT Enterprise or9In ChatGPT Enterprise and Edu workspaces, an admin must allow **Site

10 Edu workspaces.10 annotation customizations** through workspace policy.

11 11 

12[Browser annotations](https://learn.chatgpt.com/docs/browser?surface=app#app-comment-on-the-page) work on your site without any code changes.12[Browser annotations](https://learn.chatgpt.com/docs/browser?surface=app#app-comment-on-the-page) work on your site without any code changes.

13People can select part of a page, add a comment, and send it in Context to Codex or ChatGPT Work.13People can select part of a page, add a comment, and send it in Context to Codex or ChatGPT Work.

config-basic.md +7 −3

Details

95 95 

96#### Windows sandbox mode96#### Windows sandbox mode

97 97 

98When running Codex natively on Windows, set the native sandbox mode to `elevated` in the `windows` table. Use `unelevated` only if you don't have administrator permissions or if elevated setup fails.98On Windows, prefer MXC when the device and policy support it. Set a legacy implementation for fallback:

99 99 

100```toml100```toml

101[windows]101[windows]

102sandbox = "elevated" # Recommended102sandbox = "elevated" # Legacy fallback

103# sandbox = "unelevated" # Fallback if admin permissions/setup are unavailable103 

104[features]

105prefer_mxc = true

104```106```

105 107 

108Use `unelevated` only when elevated setup is unavailable and your organization's policy permits it. See the [Windows sandbox guide](https://learn.chatgpt.com/docs/windows/windows-sandbox#prefer-mxc-with-legacy-fallback) for compatibility limits and rollout controls.

109 

106#### Web search mode110#### Web search mode

107 111 

108Codex enables web search by default for local chats and serves results from a web search cache. The cache is an OpenAI-maintained index of web results, so cached mode returns pre-indexed results instead of fetching live pages. This reduces exposure to prompt injection from arbitrary live content. Treat web results as untrusted. If you are using `--yolo` or another [full access sandbox setting](https://learn.chatgpt.com/docs/agent-approvals-security#common-sandbox-and-approval-combinations), web search defaults to live results. Choose a mode with `web_search`:112Codex enables web search by default for local chats and serves results from a web search cache. The cache is an OpenAI-maintained index of web results, so cached mode returns pre-indexed results instead of fetching live pages. This reduces exposure to prompt injection from arbitrary live content. Treat web results as untrusted. If you are using `--yolo` or another [full access sandbox setting](https://learn.chatgpt.com/docs/agent-approvals-security#common-sandbox-and-approval-combinations), web search defaults to live results. Choose a mode with `web_search`:

Details

95 95 

96#### Windows sandbox mode96#### Windows sandbox mode

97 97 

98When running Codex natively on Windows, set the native sandbox mode to `elevated` in the `windows` table. Use `unelevated` only if you don't have administrator permissions or if elevated setup fails.98On Windows, prefer MXC when the device and policy support it. Set a legacy implementation for fallback:

99 99 

100```toml100```toml

101[windows]101[windows]

102sandbox = "elevated" # Recommended102sandbox = "elevated" # Legacy fallback

103# sandbox = "unelevated" # Fallback if admin permissions/setup are unavailable103 

104[features]

105prefer_mxc = true

104```106```

105 107 

108Use `unelevated` only when elevated setup is unavailable and your organization's policy permits it. See the [Windows sandbox guide](https://learn.chatgpt.com/docs/windows/windows-sandbox#prefer-mxc-with-legacy-fallback) for compatibility limits and rollout controls.

109 

106#### Web search mode110#### Web search mode

107 111 

108Codex enables web search by default for local chats and serves results from a web search cache. The cache is an OpenAI-maintained index of web results, so cached mode returns pre-indexed results instead of fetching live pages. This reduces exposure to prompt injection from arbitrary live content. Treat web results as untrusted. If you are using `--yolo` or another [full access sandbox setting](https://learn.chatgpt.com/docs/agent-approvals-security#common-sandbox-and-approval-combinations), web search defaults to live results. Choose a mode with `web_search`:112Codex enables web search by default for local chats and serves results from a web search cache. The cache is an OpenAI-maintained index of web results, so cached mode returns pre-indexed results instead of fetching live pages. This reduces exposure to prompt injection from arbitrary live content. Treat web results as untrusted. If you are using `--yolo` or another [full access sandbox setting](https://learn.chatgpt.com/docs/agent-approvals-security#common-sandbox-and-approval-combinations), web search defaults to live results. Choose a mode with `web_search`:

Details

243 },243 },

244 {244 {

245 key: "windows.sandbox",245 key: "windows.sandbox",

246 type: "unelevated | elevated | mxc",246 type: "mxc | elevated | unelevated",

247 description:247 description:

248 "Windows-only native sandbox mode when running Codex natively on Windows.",248 "Native Windows sandbox implementation. Explicit `mxc` selection fails when MXC is unavailable or prohibited by managed requirements. Use `features.prefer_mxc` with a legacy selection for automatic MXC selection with fallback.",

249 },249 },

250 {250 {

251 key: "browser_use.allow_history_access",251 key: "browser_use.allow_history_access",


865 type: "string",865 type: "string",

866 description: "Optional model override for global memory consolidation.",866 description: "Optional model override for global memory consolidation.",

867 },867 },

868 {

869 key: "features.prefer_mxc",

870 type: "boolean",

871 description:

872 "Prefer MXC for local Windows execution when native capabilities and policy allow it; otherwise retain the configured legacy sandbox and setup. Disabled by default in the standalone CLI; the desktop app can enable it through rollout configuration. Command failures don't trigger fallback.",

873 },

868 {874 {

869 key: "features.unified_exec",875 key: "features.unified_exec",

870 type: "boolean",876 type: "boolean",


1843 1849 

1844Some managed requirements enforce an exact configuration value instead of an1850Some managed requirements enforce an exact configuration value instead of an

1845allowlist. Users can't override an enforced path, update preference, login-shell1851allowlist. Users can't override an enforced path, update preference, login-shell

1846policy, feedback setting, or Windows private-desktop setting.1852policy, or feedback setting.

1847 1853 

1848Managed permission-profile allowlists require Codex 0.138.0 or later. Codex1854Managed permission-profile allowlists require Codex 0.138.0 or later. Codex

18490.137.0 and earlier ignore `allowed_permission_profiles` and managed18550.137.0 and earlier ignore `allowed_permission_profiles` and managed


2083 description:2089 description:

2084 "Allowed legacy native Windows sandbox implementations (`elevated` and `unelevated`). The list must not be empty. When both are allowed and no mode is selected, Codex prefers `elevated`. This list does not restrict the `mxc` sandbox when it is available.",2090 "Allowed legacy native Windows sandbox implementations (`elevated` and `unelevated`). The list must not be empty. When both are allowed and no mode is selected, Codex prefers `elevated`. This list does not restrict the `mxc` sandbox when it is available.",

2085 },2091 },

2092 {

2093 key: "windows.allow_mxc",

2094 type: "boolean",

2095 description:

2096 "Set to `false` to prohibit both explicit MXC selection and automatic selection through `features.prefer_mxc`. Omitting this requirement or setting it to `true` permits MXC but doesn't enable it or require it. Legacy implementation restrictions still apply to fallback.",

2097 },

2086 {2098 {

2087 key: "remote_sandbox_config",2099 key: "remote_sandbox_config",

2088 type: "array<table>",2100 type: "array<table>",

Details

719 719 

720# Leave this table empty to accept defaults. Set explicit booleans to opt in/out.720# Leave this table empty to accept defaults. Set explicit booleans to opt in/out.

721 721 

722# Prefer MXC on Windows when the device and policy support it; otherwise use the legacy sandbox below.

723 

724# Disabled by default in the standalone CLI.

725 

726prefer_mxc = true

727 

722# shell_tool = true728# shell_tool = true

723 729 

724# apps = true730# apps = true


1209 1215 

1210[windows]1216[windows]

1211 1217 

1212# Native Windows sandbox mode (Windows only): unelevated | elevated1218# Windows sandbox implementation: mxc | elevated | unelevated

1219 

1220# Legacy fallback when features.prefer_mxc is enabled. Explicit mxc selection does not fall back.

1213 1221 

1214sandbox = "unelevated"1222sandbox = "elevated"

1215```1223```

Details

243 },243 },

244 {244 {

245 key: "windows.sandbox",245 key: "windows.sandbox",

246 type: "unelevated | elevated | mxc",246 type: "mxc | elevated | unelevated",

247 description:247 description:

248 "Windows-only native sandbox mode when running Codex natively on Windows.",248 "Native Windows sandbox implementation. Explicit `mxc` selection fails when MXC is unavailable or prohibited by managed requirements. Use `features.prefer_mxc` with a legacy selection for automatic MXC selection with fallback.",

249 },249 },

250 {250 {

251 key: "browser_use.allow_history_access",251 key: "browser_use.allow_history_access",


865 type: "string",865 type: "string",

866 description: "Optional model override for global memory consolidation.",866 description: "Optional model override for global memory consolidation.",

867 },867 },

868 {

869 key: "features.prefer_mxc",

870 type: "boolean",

871 description:

872 "Prefer MXC for local Windows execution when native capabilities and policy allow it; otherwise retain the configured legacy sandbox and setup. Disabled by default in the standalone CLI; the desktop app can enable it through rollout configuration. Command failures don't trigger fallback.",

873 },

868 {874 {

869 key: "features.unified_exec",875 key: "features.unified_exec",

870 type: "boolean",876 type: "boolean",


1843 1849 

1844Some managed requirements enforce an exact configuration value instead of an1850Some managed requirements enforce an exact configuration value instead of an

1845allowlist. Users can't override an enforced path, update preference, login-shell1851allowlist. Users can't override an enforced path, update preference, login-shell

1846policy, feedback setting, or Windows private-desktop setting.1852policy, or feedback setting.

1847 1853 

1848Managed permission-profile allowlists require Codex 0.138.0 or later. Codex1854Managed permission-profile allowlists require Codex 0.138.0 or later. Codex

18490.137.0 and earlier ignore `allowed_permission_profiles` and managed18550.137.0 and earlier ignore `allowed_permission_profiles` and managed


2083 description:2089 description:

2084 "Allowed legacy native Windows sandbox implementations (`elevated` and `unelevated`). The list must not be empty. When both are allowed and no mode is selected, Codex prefers `elevated`. This list does not restrict the `mxc` sandbox when it is available.",2090 "Allowed legacy native Windows sandbox implementations (`elevated` and `unelevated`). The list must not be empty. When both are allowed and no mode is selected, Codex prefers `elevated`. This list does not restrict the `mxc` sandbox when it is available.",

2085 },2091 },

2092 {

2093 key: "windows.allow_mxc",

2094 type: "boolean",

2095 description:

2096 "Set to `false` to prohibit both explicit MXC selection and automatic selection through `features.prefer_mxc`. Omitting this requirement or setting it to `true` permits MXC but doesn't enable it or require it. Legacy implementation restrictions still apply to fallback.",

2097 },

2086 {2098 {

2087 key: "remote_sandbox_config",2099 key: "remote_sandbox_config",

2088 type: "array<table>",2100 type: "array<table>",

config-sample.md +10 −2

Details

719 719 

720# Leave this table empty to accept defaults. Set explicit booleans to opt in/out.720# Leave this table empty to accept defaults. Set explicit booleans to opt in/out.

721 721 

722# Prefer MXC on Windows when the device and policy support it; otherwise use the legacy sandbox below.

723 

724# Disabled by default in the standalone CLI.

725 

726prefer_mxc = true

727 

722# shell_tool = true728# shell_tool = true

723 729 

724# apps = true730# apps = true


1209 1215 

1210[windows]1216[windows]

1211 1217 

1212# Native Windows sandbox mode (Windows only): unelevated | elevated1218# Windows sandbox implementation: mxc | elevated | unelevated

1219 

1220# Legacy fallback when features.prefer_mxc is enabled. Explicit mxc selection does not fall back.

1213 1221 

1214sandbox = "unelevated"1222sandbox = "elevated"

1215```1223```

Details

4 4 

5## Why you would want this5## Why you would want this

6 6 

7If you already have a model gateway and use Codex with the API Platform, you can keep your gateway while moving to **Sign in with ChatGPT**. This gives you access to eligible ChatGPT workspace features and controls, including [voice mode](#voice-through-a-gateway), while keeping your model gateway. Compare the sign-in methods in [Feature availability](https://learn.chatgpt.com/docs/pricing#feature-availability).7{/* vale Microsoft.Foreign = NO */}

8 

9If you already have a model gateway and use Codex with the API Platform, you can keep your gateway while moving to **Sign in with ChatGPT**. This gives you access to eligible ChatGPT workspace features and controls (e.g. [**voice mode**](#voice-through-a-gateway)) while keeping your model gateway. Compare the sign-in methods in [Feature availability](https://learn.chatgpt.com/docs/pricing#feature-availability).

10 

11{/* vale Microsoft.Foreign = YES */}

8 12 

9## How to set it up13## How to set it up

10 14 


61codex_with_gateway65codex_with_gateway

62```66```

63 67 

64Configure your gateway to accept `X-Gateway-Key`, validate the key, and remove that header before forwarding the request. The request now carries:68Configure your gateway to accept `X-Gateway-Key`, validate the key, and remove that header before forwarding the request to the backend. The request from the client now carries:

65 69 

66```http70```http

67Authorization: Bearer <ChatGPT token>71Authorization: Bearer <ChatGPT token>

models.md +5 −3

Details

165GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna remain available during the rollout. Selecting a165GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna remain available during the rollout. Selecting a

166new model doesn't change workspace permissions or grant access to it.166new model doesn't change workspace permissions or grant access to it.

167 167 

168For Astra Ultrafast availability and credit usage on Pro $500 and eligible168For Ultrafast mode availability and credit usage on Pro $500 and eligible

169Enterprise and Edu plans, see [Speed](https://learn.chatgpt.com/docs/agent-configuration/speed#ultrafast-mode).169Enterprise and Edu plans, see [Speed](https://learn.chatgpt.com/docs/agent-configuration/speed#ultrafast-mode).

170 170 

171<a id="app-compare-models"></a>171<a id="app-compare-models"></a>


295administrator enables it.295administrator enables it.

296Free and Go are not included at launch.296Free and Go are not included at launch.

297 297 

298Standard and Fast modes are available at launch. Ultrafast support for298Standard and Fast modes are available where supported. Ultrafast is available

299GPT-6.1 Sol is coming later.299on Pro $500 and eligible Enterprise and Edu plans. See

300[Ultrafast mode](https://learn.chatgpt.com/docs/agent-configuration/speed#ultrafast-mode) for plan and

301workspace requirements.

300 302 

301Use `gpt-6.1-sol`. Available controls depend on your plan, client, and workspace303Use `gpt-6.1-sol`. Available controls depend on your plan, client, and workspace

302settings.304settings.

permissions.md +5 −4

Details

517 enforcement path depends on user namespaces and kernel support; restricted517 enforcement path depends on user namespaces and kernel support; restricted

518 container hosts can force compatibility paths, and unsupported split policies518 container hosts can force compatibility paths, and unsupported split policies

519 are refused.519 are refused.

520- On native Windows, [`elevated` sandboxing](https://learn.chatgpt.com/docs/windows/windows-sandbox#windows-sandbox)520- On Windows, MXC uses process isolation for filesystem and network permissions.

521 is strongest because it can use dedicated lower-privilege sandbox users,521 Review [MXC compatibility](https://learn.chatgpt.com/docs/windows/windows-sandbox#mxc-compatibility)

522 filesystem permission boundaries, and firewall rules. `unelevated`522 before enabling it. The legacy `elevated` fallback uses dedicated lower-privilege

523 sandboxing is a fallback with weaker network isolation and cannot enforce523 sandbox users, filesystem permission boundaries, and firewall rules.

524 `unelevated` is a legacy fallback with weaker network isolation and cannot enforce

524 every split read/write carveout, so unsupported policies are refused. Use WSL525 every split read/write carveout, so unsupported policies are refused. Use WSL

525 when you need the Linux sandbox model.526 when you need the Linux sandbox model.

526 527 

speed.md +10 −15

Details

17 17 

18For supported models, Fast mode uses included subscription limits at 2.5x the18For supported models, Fast mode uses included subscription limits at 2.5x the

19Standard rate. Purchased credits and Enterprise pay-as-you-go usage are billed19Standard rate. Purchased credits and Enterprise pay-as-you-go usage are billed

20at 2x the Standard rate. These billing multipliers don't describe speed increases.20at 2x the Standard rate.

21 21 

22<a id="astra-ultrafast"></a>22<a id="astra-ultrafast"></a>

23 23 

24## Ultrafast mode24## Ultrafast mode

25 25 

26Ultrafast supports GPT-6 Astra and GPT-6.1 Sol.

27 

26GPT-6 Astra Ultrafast generates tokens up to 8x faster than GPT-6 Astra in28GPT-6 Astra Ultrafast generates tokens up to 8x faster than GPT-6 Astra in

27Standard mode in Codex. This comparison measures token generation speed,29Standard mode in Codex.

28not billing rates or overall task completion time.

29 30 

30Ultrafast is available in Codex and ChatGPT Work on Pro $500 and eligible31Ultrafast is available in Codex and ChatGPT Work on Pro $500 and eligible

31Enterprise and Edu plans. On Pro $500, Ultrafast uses your included usage first,32Enterprise and Edu plans. On Pro $500, Ultrafast uses your included usage first,

32then your available credits after that allowance runs out.33then your available credits after that allowance runs out.

33 34 

34For GPT-6 Astra, Ultrafast uses included subscription limits at 8x the35Ultrafast mode bills purchased credits and Enterprise pay-as-you-go usage at

35Standard rate. Purchased credits and Enterprise pay-as-you-go usage are billed366x the Standard rate. Ultrafast uses included subscription limits at 8x the

36at 6x the Standard rate. These billing multipliers don't describe speed increases.37Standard rate.

37Enterprise billing remains subject to the workspace's agreement.

38 38 

39For Enterprise workspaces, Ultrafast is off by default. Workspace owners can39For Enterprise workspaces, Ultrafast is off by default. Workspace owners can

40enable access for selected users or the workspace through40enable access for selected users or the workspace through


48 pricing, credits, and usage limits. See [Codex pricing](https://learn.chatgpt.com/docs/pricing) for48 pricing, credits, and usage limits. See [Codex pricing](https://learn.chatgpt.com/docs/pricing) for

49 details.49 details.

50 50 

51### Supported models

52 

53GPT-6 Astra, GPT-6 Sol, and GPT-6 Luna support Fast mode where available.

54[GPT-6.1 Sol](https://learn.chatgpt.com/docs/models#gpt-61-sol) supports Standard and Fast where

55available. Access depends on your plan, client, workspace settings, and rollout.

56See [Models](https://learn.chatgpt.com/docs/models) for model availability and51See [Models](https://learn.chatgpt.com/docs/models) for model availability and

57[Pricing](https://learn.chatgpt.com/docs/pricing#token-rates) for token rates.52[Pricing](https://learn.chatgpt.com/docs/pricing#token-rates) for token rates.

58 53 


66Other self-serve plans don't have access to Ultrafast at launch, even with61Other self-serve plans don't have access to Ultrafast at launch, even with

67purchased credits.62purchased credits.

68 63 

69Ultrafast isn't available to workspaces that require inference residency64GPT-6.1 Sol Ultrafast supports inference residency in the United States and

70outside the United States. A workspace's location alone doesn't determine65Europe (EEA + Switzerland). GPT-6 Astra Ultrafast supports inference residency

71eligibility.66in the United States only.

72 67 

73### API billing68### API billing

74 69 

whats-new.md +10 −0

Details

4 4 

5For every versioned update, bug fix, and minor improvement, see the [Codex changelog](https://learn.chatgpt.com/docs/changelog) and [API changelog](https://developers.openai.com/api/docs/changelog).5For every versioned update, bug fix, and minor improvement, see the [Codex changelog](https://learn.chatgpt.com/docs/changelog) and [API changelog](https://developers.openai.com/api/docs/changelog).

6 6 

7## [October 5–9, 2026](https://learn.chatgpt.com/docs/whats-new/october-5-9-2026)

8 

9### Use GPT-6.1 Sol with Ultrafast mode

10 

11[Ultrafast mode](https://learn.chatgpt.com/docs/agent-configuration/speed#ultrafast-mode) speeds up

12token generation with GPT-6.1 Sol in Codex and ChatGPT Work. Ultrafast is

13available on Pro $500 and eligible Enterprise and Edu plans.

14 

15[Read the full update →](https://learn.chatgpt.com/docs/whats-new/october-5-9-2026)

16 

7## [DevDay 2026](https://learn.chatgpt.com/docs/whats-new/devday-2026)17## [DevDay 2026](https://learn.chatgpt.com/docs/whats-new/devday-2026)

8 18 

9<a id="devday-roundup-title" />19<a id="devday-roundup-title" />

Details

1# October 5–9, 2026

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[Back to What's new](https://learn.chatgpt.com/docs/whats-new#october-59-2026)

6 

7## Use GPT-6.1 Sol with Ultrafast mode

8 

9Ultrafast mode speeds up token generation with [GPT-6.1 Sol](https://learn.chatgpt.com/docs/models#gpt-61-sol)

10in Codex and ChatGPT Work. Ultrafast is available on Pro $500 and eligible

11Enterprise and Edu plans.

12 

13Enterprise access is off by default; workspace owners must enable it.

14GPT-6.1 Sol Ultrafast supports inference residency in the United States and

15Europe (EEA + Switzerland).

16 

17See [Ultrafast mode](https://learn.chatgpt.com/docs/agent-configuration/speed#ultrafast-mode) for plan

18and workspace requirements and [Pricing](https://learn.chatgpt.com/docs/pricing) for usage and credit

19rates. Read the [launch announcement](https://learn.chatgpt.com/docs/changelog#codex-2026-10-08-gpt-61-sol-ultrafast).

windows.md +160 −48

Details

32 32 

33 33 

34 34 

35The native Windows sandbox has two modes:35Codex supports three Windows sandbox implementations:

36 36 

37- natively on Windows with the stronger `elevated` sandbox,37- `mxc`: The recommended sandbox on compatible Windows devices where policy permits it. Uses process isolation without administrator-approved setup, additional Windows accounts, changes to host file permissions, or local firewall rules.

38- natively on Windows with the fallback `unelevated` sandbox.38- `elevated`: The preferred legacy fallback when MXC is unavailable or disabled. Requires administrator-approved setup. Commands in the sandbox run without administrator privileges.

39- `unelevated`: A legacy fallback when elevated setup is unavailable and organizational policy permits it. Has weaker network isolation than `elevated` and doesn't support denied read paths.

39 40 

40 41 

41 42 


43 44 

44## Configure the Windows sandbox45## Configure the Windows sandbox

45 46 

46When you run Codex natively on Windows, agent mode uses a Windows sandbox to47The Windows sandbox enforces the active filesystem and network permissions for

47block filesystem writes outside the working folder and prevent network access48commands and their child processes. The permission profile determines which

48without your explicit approval.49paths are readable or writable and whether network access is allowed. Approval

50policy separately controls when Codex asks to run commands with more access.

51See [sandbox and approvals](https://learn.chatgpt.com/docs/agent-approvals-security).

49 52 

50Native Windows sandbox support includes two modes that you can configure in53### Control MXC rollout

51`config.toml`:54 

55These settings are available in Codex CLI 0.162.0. In the standalone CLI,

56`features.prefer_mxc` is off by default. The desktop app can enable this

57preference through its rollout configuration.

58 

59#### Prefer MXC with legacy fallback

60 

61To use MXC when the device and policy support it, add this to `config.toml`:

62 

63```toml

64[features]

65prefer_mxc = true

66```

67 

68Keep `windows.sandbox` set to your organization's permitted legacy implementation,

69`elevated` or `unelevated`, for fallback.

70 

71Codex uses MXC for local Windows commands when the device and policy support it,

72even when `windows.sandbox` selects a legacy implementation. Otherwise, it uses

73the existing legacy selection and setup flow. This includes devices without

74MXC support and policies that set `windows.allow_mxc = false` or forbid local

75binding. `windows.allowed_sandbox_implementations` still constrains fallback;

76permission profiles and other managed requirements continue to apply.

77 

78Fallback happens during sandbox selection. Commands that fail after MXC is

79selected aren't retried in a legacy sandbox.

80 

81Administrators can distribute this configuration as a default or enforce

82`features.prefer_mxc = true` through `requirements.toml`. Both permit legacy

83fallback. See [managed configuration](https://learn.chatgpt.com/docs/enterprise/managed-configuration)

84for how defaults and requirements differ.

85 

86#### Keep MXC disabled

87 

88To prevent MXC use in your organization, add this to your managed

89`requirements.toml`:

90 

91```toml

92[windows]

93allow_mxc = false

94```

95 

96This blocks both automatic MXC selection and explicit `windows.sandbox = "mxc"`.

97Existing legacy sandbox settings and requirements still apply. Configure

98`windows.allow_mxc` in `requirements.toml`, not `config.toml`.

99 

100### MXC compatibility

101 

102Microsoft Execution Containers (MXC) uses native Windows process isolation

103without creating sandbox accounts, changing host file permissions, or running

104the classic elevated setup. Commands run under the user's Windows identity

105with a policy applied to each command. MXC doesn't require administrator

106elevation for sandbox setup or local Windows Firewall rules.

107MXC accepts readable, writable, and denied paths from the active permission

108profile. Its native network policy controls command network access without

109depending on the legacy sandbox's firewall provisioning.

110 

111Before selecting MXC, validate the required capabilities and your workloads on

112the Windows 11 device. A Windows version number alone doesn't establish

113compatibility.

114 

115In Codex CLI 0.162.0, you can test MXC for one command without changing the saved

116sandbox selection. From your project directory, run:

117 

118```powershell

119codex -c windows.sandbox=mxc sandbox --include-managed-config --permission-profile :workspace -- cmd.exe /d /c echo MXC_OK

120$LASTEXITCODE

121```

122 

123Expected output is `MXC_OK` and exit code `0`. This checks command startup with

124the workspace permission profile and managed requirements. Also test permitted

125file access, expected denials, PowerShell, and your network policy before using

126MXC for normal work.

127 

128Explicit `windows.sandbox = "mxc"` selection fails if the required native

129capabilities are unavailable; it doesn't fall back to a classic implementation.

130Policies with denied paths also require native deny-path support.

131 

132Check these compatibility limits:

133 

134- Managed networking requires effective `allow_local_binding = true`. MXC

135 permits connections to and from services on host loopback. Proxy domain rules

136 still apply to proxied traffic, but the proxy's additional private-network

137 destination checks are removed. This doesn't enable networking when it's disabled.

138- Remaining child processes stop when the foreground command exits. Test

139 workflows that rely on detached development servers.

140 

141### Configure a legacy fallback

142 

143Select the fallback implementation in `config.toml`:

52 144 

53```toml145```toml

54[windows]146[windows]

55sandbox = "elevated" # or "unelevated"147sandbox = "elevated" # or "unelevated"

56```148```

57 149 

58`elevated` is the preferred native Windows sandbox. It uses dedicated150`elevated` is the preferred legacy fallback. It uses dedicated

59lower-privilege sandbox users, filesystem permission boundaries, firewall151lower-privilege sandbox users, filesystem permission boundaries, firewall

60rules, and local policy changes needed for commands that run in the sandbox.152rules, and local policy changes needed for commands that run in the sandbox.

61 153 

62`unelevated` is the fallback native Windows sandbox. It runs commands with a154`unelevated` is a legacy fallback. It runs commands with a

63restricted Windows token derived from your current user, applies ACL-based155restricted Windows token derived from your current user, applies ACL-based

64filesystem boundaries, and uses environment-level offline controls instead of156filesystem boundaries, and uses environment-level offline controls instead of

65the dedicated offline-user firewall rule. It's weaker than `elevated`, but it157the dedicated offline-user firewall rule. It provides weaker network isolation

66is still useful when administrator-approved setup is blocked by local or158than `elevated` and doesn't support denied read paths, but is still useful when

67enterprise policy.159administrator-approved setup is blocked by local or enterprise policy.

68 160 

69If both modes are available, use `elevated`. If the default native sandbox161Use MXC when the device and policy support it. Otherwise, prefer `elevated`.

70doesn't work in your environment, use `unelevated` as a fallback while you162Use `unelevated` as a fallback only when your organization's policy permits it.

71troubleshoot the setup.

72 163 

73Enterprise administrators can constrain which native sandbox implementations164Enterprise administrators can constrain which classic sandbox implementations

74Codex can use through [`requirements.toml`](https://learn.chatgpt.com/docs/enterprise/managed-configuration#admin-enforced-requirements-requirementstoml):165Codex can use through [`requirements.toml`](https://learn.chatgpt.com/docs/enterprise/managed-configuration#admin-enforced-requirements-requirementstoml):

75 166 

76```toml167```toml


78allowed_sandbox_implementations = ["elevated"]169allowed_sandbox_implementations = ["elevated"]

79```170```

80 171 

81This example requires the `elevated` sandbox and prevents users from falling172This example permits `elevated` and prevents fallback to `unelevated`. It does

82back to `unelevated`. To permit either implementation, include both values;173not restrict `mxc` when MXC is available. Other managed permission and network

83Codex prefers `elevated` when no mode is selected. See the174requirements still apply. To permit either classic implementation, include

175both values; Codex prefers `elevated` when no mode is selected. See the

84[`requirements.toml` reference](https://learn.chatgpt.com/docs/config-file/config-reference#requirementstoml) for176[`requirements.toml` reference](https://learn.chatgpt.com/docs/config-file/config-reference#requirementstoml) for

85the supported values.177the supported values. To block MXC as well, use the separate

178[`windows.allow_mxc` requirement](#keep-mxc-disabled).

179 

180By default, both legacy sandbox modes also use a private desktop for stronger UI

181isolation.

182 

183#### Provision the classic elevated sandbox

86 184 

87By default, both sandbox modes also use a private desktop for stronger UI185For employees without local administrator rights, IT can install the CLI and

88isolation. Set `windows.sandbox_private_desktop = false` only if you need the186provision the sandbox before the employee starts Codex. From an elevated

89older `Winsta0\\Default` behavior for compatibility.187deployment process, run:

188 

189```powershell

190codex sandbox setup --elevated --user 'DOMAIN\alice' --codex-home 'C:\Users\alice\.codex'

191```

192 

193Replace the identity and path with the employee's Windows identity and

194`CODEX_HOME`. The command reads that user's configuration, provisions the

195sandbox, and saves `windows.sandbox = "elevated"`. The employee then runs Codex

196from a normal terminal. Using a non-admin terminal doesn't select the

197`unelevated` implementation.

90 198 

91### Sandbox permissions199### Sandbox permissions

92 200 


111 219 

112- `winget` should be available. If it's missing, update Windows or install220- `winget` should be available. If it's missing, update Windows or install

113 the Windows Package Manager before setting up Codex.221 the Windows Package Manager before setting up Codex.

114- The recommended native sandbox depends on administrator-approved setup.222- The classic `elevated` sandbox depends on administrator-approved setup.

115- Some enterprise-managed devices block the required setup steps even when the223- Some enterprise-managed devices block the required setup steps even when the

116 OS version itself is acceptable.224 OS version itself is acceptable.

225- MXC additionally requires the native capabilities described in

226 [MXC compatibility](#mxc-compatibility); this matrix doesn't establish MXC

227 availability on a particular device.

117 228 

118### Grant sandbox read access229### Check sandbox read access

119 

120When a command fails because the Windows sandbox can't read a directory, use:

121 

122```text

123/sandbox-add-read-dir C:\absolute\directory\path

124```

125 230 

126The path must be an existing absolute directory. After the command succeeds, later commands that run in the sandbox can read that directory during the current session.231When a command can't read a directory, check the active permission profile,

232managed requirements, and Windows file permissions. In the CLI, use `/status`

233and `/debug-config` to inspect the active session and configuration. Ask your

234administrator to review a managed restriction rather than disabling the sandbox.

127 235 

128 236 

129 237 


131 239 

132Use the native Windows sandbox by default. Choose [WSL](https://learn.chatgpt.com/docs/windows/wsl)240Use the native Windows sandbox by default. Choose [WSL](https://learn.chatgpt.com/docs/windows/wsl)

133when you need Linux-native tooling, your workflow already lives in WSL2, or241when you need Linux-native tooling, your workflow already lives in WSL2, or

134neither native Windows sandbox mode meets your needs.242the available native Windows implementations don't meet your needs.

135 243 

136## Troubleshooting and FAQ244## Troubleshooting and FAQ

137 245 

138If you are troubleshooting a managed Windows machine, start with the native246If you are troubleshooting a managed Windows machine, start with the native

139sandbox mode, Windows version, and any policy error shown by Codex. Most native247sandbox mode, Windows version, and any policy error shown by Codex. For MXC,

140Windows support issues come from sandbox setup, logon rights, or filesystem248check [compatibility](#mxc-compatibility) and the effective network policy.

141permissions rather than from the editor itself.249Legacy sandbox issues can come from setup, logon rights, or filesystem permissions.

142 250 

143 251 

144 252 

145### My native sandbox setup failed253### My legacy sandbox setup failed

146 254 

147 255 

148If Codex cannot complete the `elevated` sandbox setup, the most common causes256If Codex cannot complete the `elevated` sandbox setup, the most common causes


1612. If your company laptop blocks this, ask your IT team whether the machine2692. If your company laptop blocks this, ask your IT team whether the machine

162 allows administrator-approved setup for local user/group creation, firewall270 allows administrator-approved setup for local user/group creation, firewall

163 configuration, and the required sandbox-user logon rights.271 configuration, and the required sandbox-user logon rights.

1643. If the default setup still fails, use the `unelevated` sandbox so you can2723. If setup still fails and managed policy permits it, use `unelevated` while

165 continue working while the issue is investigated.273 the issue is investigated.

166 274 

167 275 

168 276 


173### Codex switched me to the unelevated sandbox281### Codex switched me to the unelevated sandbox

174 282 

175 283 

176This means Codex could not finish the stronger `elevated` sandbox setup on your284The `unelevated` implementation may be selected in configuration or used as a

177machine.285fallback when `elevated` setup isn't available.

178 286 

179- Codex can still run in a sandboxed mode.287- Codex can still run in a sandboxed mode.

180- It still applies ACL-based filesystem boundaries, but it does not use the288- It still applies ACL-based filesystem boundaries, but it does not use the


183- This is a useful fallback, but not the preferred long-term enterprise291- This is a useful fallback, but not the preferred long-term enterprise

184 configuration.292 configuration.

185 293 

186If you are on a managed enterprise laptop, the best long-term fix is usually to294For a managed enterprise laptop, check [MXC compatibility](#mxc-compatibility)

187get the `elevated` sandbox working with help from your IT team.295first. If MXC is unavailable or disabled, ask your IT team to provision

296`elevated`.

188 297 

189 298 

190 299 


275What to try:384What to try:

276 385 

2771. Restart Codex.3861. Restart Codex.

2782. Try the `elevated` sandbox setup again.3872. For MXC, repeat the [compatibility probe](#mxc-compatibility) and check the

2793. If that does not fix it, use the `unelevated` sandbox as a temporary388 effective network policy. For the classic `elevated` implementation, try

280 fallback.389 sandbox setup again.

3903. If a classic sandbox is needed and managed policy permits it, use

391 `unelevated` as a temporary fallback.

2814. Collect the sandbox log for review.3924. Collect the sandbox log for review.

282 393 

283 394 


296It is also helpful to include:407It is also helpful to include:

297 408 

298- a short description of what you were trying to do,409- a short description of what you were trying to do,

299- whether the `elevated` sandbox failed or the `unelevated` sandbox was used,410- the selected implementation: `mxc`, `elevated`, or `unelevated`,

300- any error message shown in the app,411- any error message shown in the app,

301- whether you saw `1385` or another Windows or PowerShell error,412- whether you saw `1385` or another Windows or PowerShell error,

413- your Windows build number,

302- and whether you are on Windows 11 or Windows 10.414- and whether you are on Windows 11 or Windows 10.

303 415 

304Do not send:416Do not send:

Details

32 32 

33 33 

34 34 

35The native Windows sandbox has two modes:35Codex supports three Windows sandbox implementations:

36 36 

37- natively on Windows with the stronger `elevated` sandbox,37- `mxc`: The recommended sandbox on compatible Windows devices where policy permits it. Uses process isolation without administrator-approved setup, additional Windows accounts, changes to host file permissions, or local firewall rules.

38- natively on Windows with the fallback `unelevated` sandbox.38- `elevated`: The preferred legacy fallback when MXC is unavailable or disabled. Requires administrator-approved setup. Commands in the sandbox run without administrator privileges.

39- `unelevated`: A legacy fallback when elevated setup is unavailable and organizational policy permits it. Has weaker network isolation than `elevated` and doesn't support denied read paths.

39 40 

40 41 

41 42 


43 44 

44## Configure the Windows sandbox45## Configure the Windows sandbox

45 46 

46When you run Codex natively on Windows, agent mode uses a Windows sandbox to47The Windows sandbox enforces the active filesystem and network permissions for

47block filesystem writes outside the working folder and prevent network access48commands and their child processes. The permission profile determines which

48without your explicit approval.49paths are readable or writable and whether network access is allowed. Approval

50policy separately controls when Codex asks to run commands with more access.

51See [sandbox and approvals](https://learn.chatgpt.com/docs/agent-approvals-security).

49 52 

50Native Windows sandbox support includes two modes that you can configure in53### Control MXC rollout

51`config.toml`:54 

55These settings are available in Codex CLI 0.162.0. In the standalone CLI,

56`features.prefer_mxc` is off by default. The desktop app can enable this

57preference through its rollout configuration.

58 

59#### Prefer MXC with legacy fallback

60 

61To use MXC when the device and policy support it, add this to `config.toml`:

62 

63```toml

64[features]

65prefer_mxc = true

66```

67 

68Keep `windows.sandbox` set to your organization's permitted legacy implementation,

69`elevated` or `unelevated`, for fallback.

70 

71Codex uses MXC for local Windows commands when the device and policy support it,

72even when `windows.sandbox` selects a legacy implementation. Otherwise, it uses

73the existing legacy selection and setup flow. This includes devices without

74MXC support and policies that set `windows.allow_mxc = false` or forbid local

75binding. `windows.allowed_sandbox_implementations` still constrains fallback;

76permission profiles and other managed requirements continue to apply.

77 

78Fallback happens during sandbox selection. Commands that fail after MXC is

79selected aren't retried in a legacy sandbox.

80 

81Administrators can distribute this configuration as a default or enforce

82`features.prefer_mxc = true` through `requirements.toml`. Both permit legacy

83fallback. See [managed configuration](https://learn.chatgpt.com/docs/enterprise/managed-configuration)

84for how defaults and requirements differ.

85 

86#### Keep MXC disabled

87 

88To prevent MXC use in your organization, add this to your managed

89`requirements.toml`:

90 

91```toml

92[windows]

93allow_mxc = false

94```

95 

96This blocks both automatic MXC selection and explicit `windows.sandbox = "mxc"`.

97Existing legacy sandbox settings and requirements still apply. Configure

98`windows.allow_mxc` in `requirements.toml`, not `config.toml`.

99 

100### MXC compatibility

101 

102Microsoft Execution Containers (MXC) uses native Windows process isolation

103without creating sandbox accounts, changing host file permissions, or running

104the classic elevated setup. Commands run under the user's Windows identity

105with a policy applied to each command. MXC doesn't require administrator

106elevation for sandbox setup or local Windows Firewall rules.

107MXC accepts readable, writable, and denied paths from the active permission

108profile. Its native network policy controls command network access without

109depending on the legacy sandbox's firewall provisioning.

110 

111Before selecting MXC, validate the required capabilities and your workloads on

112the Windows 11 device. A Windows version number alone doesn't establish

113compatibility.

114 

115In Codex CLI 0.162.0, you can test MXC for one command without changing the saved

116sandbox selection. From your project directory, run:

117 

118```powershell

119codex -c windows.sandbox=mxc sandbox --include-managed-config --permission-profile :workspace -- cmd.exe /d /c echo MXC_OK

120$LASTEXITCODE

121```

122 

123Expected output is `MXC_OK` and exit code `0`. This checks command startup with

124the workspace permission profile and managed requirements. Also test permitted

125file access, expected denials, PowerShell, and your network policy before using

126MXC for normal work.

127 

128Explicit `windows.sandbox = "mxc"` selection fails if the required native

129capabilities are unavailable; it doesn't fall back to a classic implementation.

130Policies with denied paths also require native deny-path support.

131 

132Check these compatibility limits:

133 

134- Managed networking requires effective `allow_local_binding = true`. MXC

135 permits connections to and from services on host loopback. Proxy domain rules

136 still apply to proxied traffic, but the proxy's additional private-network

137 destination checks are removed. This doesn't enable networking when it's disabled.

138- Remaining child processes stop when the foreground command exits. Test

139 workflows that rely on detached development servers.

140 

141### Configure a legacy fallback

142 

143Select the fallback implementation in `config.toml`:

52 144 

53```toml145```toml

54[windows]146[windows]

55sandbox = "elevated" # or "unelevated"147sandbox = "elevated" # or "unelevated"

56```148```

57 149 

58`elevated` is the preferred native Windows sandbox. It uses dedicated150`elevated` is the preferred legacy fallback. It uses dedicated

59lower-privilege sandbox users, filesystem permission boundaries, firewall151lower-privilege sandbox users, filesystem permission boundaries, firewall

60rules, and local policy changes needed for commands that run in the sandbox.152rules, and local policy changes needed for commands that run in the sandbox.

61 153 

62`unelevated` is the fallback native Windows sandbox. It runs commands with a154`unelevated` is a legacy fallback. It runs commands with a

63restricted Windows token derived from your current user, applies ACL-based155restricted Windows token derived from your current user, applies ACL-based

64filesystem boundaries, and uses environment-level offline controls instead of156filesystem boundaries, and uses environment-level offline controls instead of

65the dedicated offline-user firewall rule. It's weaker than `elevated`, but it157the dedicated offline-user firewall rule. It provides weaker network isolation

66is still useful when administrator-approved setup is blocked by local or158than `elevated` and doesn't support denied read paths, but is still useful when

67enterprise policy.159administrator-approved setup is blocked by local or enterprise policy.

68 160 

69If both modes are available, use `elevated`. If the default native sandbox161Use MXC when the device and policy support it. Otherwise, prefer `elevated`.

70doesn't work in your environment, use `unelevated` as a fallback while you162Use `unelevated` as a fallback only when your organization's policy permits it.

71troubleshoot the setup.

72 163 

73Enterprise administrators can constrain which native sandbox implementations164Enterprise administrators can constrain which classic sandbox implementations

74Codex can use through [`requirements.toml`](https://learn.chatgpt.com/docs/enterprise/managed-configuration#admin-enforced-requirements-requirementstoml):165Codex can use through [`requirements.toml`](https://learn.chatgpt.com/docs/enterprise/managed-configuration#admin-enforced-requirements-requirementstoml):

75 166 

76```toml167```toml


78allowed_sandbox_implementations = ["elevated"]169allowed_sandbox_implementations = ["elevated"]

79```170```

80 171 

81This example requires the `elevated` sandbox and prevents users from falling172This example permits `elevated` and prevents fallback to `unelevated`. It does

82back to `unelevated`. To permit either implementation, include both values;173not restrict `mxc` when MXC is available. Other managed permission and network

83Codex prefers `elevated` when no mode is selected. See the174requirements still apply. To permit either classic implementation, include

175both values; Codex prefers `elevated` when no mode is selected. See the

84[`requirements.toml` reference](https://learn.chatgpt.com/docs/config-file/config-reference#requirementstoml) for176[`requirements.toml` reference](https://learn.chatgpt.com/docs/config-file/config-reference#requirementstoml) for

85the supported values.177the supported values. To block MXC as well, use the separate

178[`windows.allow_mxc` requirement](#keep-mxc-disabled).

179 

180By default, both legacy sandbox modes also use a private desktop for stronger UI

181isolation.

182 

183#### Provision the classic elevated sandbox

86 184 

87By default, both sandbox modes also use a private desktop for stronger UI185For employees without local administrator rights, IT can install the CLI and

88isolation. Set `windows.sandbox_private_desktop = false` only if you need the186provision the sandbox before the employee starts Codex. From an elevated

89older `Winsta0\\Default` behavior for compatibility.187deployment process, run:

188 

189```powershell

190codex sandbox setup --elevated --user 'DOMAIN\alice' --codex-home 'C:\Users\alice\.codex'

191```

192 

193Replace the identity and path with the employee's Windows identity and

194`CODEX_HOME`. The command reads that user's configuration, provisions the

195sandbox, and saves `windows.sandbox = "elevated"`. The employee then runs Codex

196from a normal terminal. Using a non-admin terminal doesn't select the

197`unelevated` implementation.

90 198 

91### Sandbox permissions199### Sandbox permissions

92 200 


111 219 

112- `winget` should be available. If it's missing, update Windows or install220- `winget` should be available. If it's missing, update Windows or install

113 the Windows Package Manager before setting up Codex.221 the Windows Package Manager before setting up Codex.

114- The recommended native sandbox depends on administrator-approved setup.222- The classic `elevated` sandbox depends on administrator-approved setup.

115- Some enterprise-managed devices block the required setup steps even when the223- Some enterprise-managed devices block the required setup steps even when the

116 OS version itself is acceptable.224 OS version itself is acceptable.

225- MXC additionally requires the native capabilities described in

226 [MXC compatibility](#mxc-compatibility); this matrix doesn't establish MXC

227 availability on a particular device.

117 228 

118### Grant sandbox read access229### Check sandbox read access

119 

120When a command fails because the Windows sandbox can't read a directory, use:

121 

122```text

123/sandbox-add-read-dir C:\absolute\directory\path

124```

125 230 

126The path must be an existing absolute directory. After the command succeeds, later commands that run in the sandbox can read that directory during the current session.231When a command can't read a directory, check the active permission profile,

232managed requirements, and Windows file permissions. In the CLI, use `/status`

233and `/debug-config` to inspect the active session and configuration. Ask your

234administrator to review a managed restriction rather than disabling the sandbox.

127 235 

128 236 

129 237 


131 239 

132Use the native Windows sandbox by default. Choose [WSL](https://learn.chatgpt.com/docs/windows/wsl)240Use the native Windows sandbox by default. Choose [WSL](https://learn.chatgpt.com/docs/windows/wsl)

133when you need Linux-native tooling, your workflow already lives in WSL2, or241when you need Linux-native tooling, your workflow already lives in WSL2, or

134neither native Windows sandbox mode meets your needs.242the available native Windows implementations don't meet your needs.

135 243 

136## Troubleshooting and FAQ244## Troubleshooting and FAQ

137 245 

138If you are troubleshooting a managed Windows machine, start with the native246If you are troubleshooting a managed Windows machine, start with the native

139sandbox mode, Windows version, and any policy error shown by Codex. Most native247sandbox mode, Windows version, and any policy error shown by Codex. For MXC,

140Windows support issues come from sandbox setup, logon rights, or filesystem248check [compatibility](#mxc-compatibility) and the effective network policy.

141permissions rather than from the editor itself.249Legacy sandbox issues can come from setup, logon rights, or filesystem permissions.

142 250 

143 251 

144 252 

145### My native sandbox setup failed253### My legacy sandbox setup failed

146 254 

147 255 

148If Codex cannot complete the `elevated` sandbox setup, the most common causes256If Codex cannot complete the `elevated` sandbox setup, the most common causes


1612. If your company laptop blocks this, ask your IT team whether the machine2692. If your company laptop blocks this, ask your IT team whether the machine

162 allows administrator-approved setup for local user/group creation, firewall270 allows administrator-approved setup for local user/group creation, firewall

163 configuration, and the required sandbox-user logon rights.271 configuration, and the required sandbox-user logon rights.

1643. If the default setup still fails, use the `unelevated` sandbox so you can2723. If setup still fails and managed policy permits it, use `unelevated` while

165 continue working while the issue is investigated.273 the issue is investigated.

166 274 

167 275 

168 276 


173### Codex switched me to the unelevated sandbox281### Codex switched me to the unelevated sandbox

174 282 

175 283 

176This means Codex could not finish the stronger `elevated` sandbox setup on your284The `unelevated` implementation may be selected in configuration or used as a

177machine.285fallback when `elevated` setup isn't available.

178 286 

179- Codex can still run in a sandboxed mode.287- Codex can still run in a sandboxed mode.

180- It still applies ACL-based filesystem boundaries, but it does not use the288- It still applies ACL-based filesystem boundaries, but it does not use the


183- This is a useful fallback, but not the preferred long-term enterprise291- This is a useful fallback, but not the preferred long-term enterprise

184 configuration.292 configuration.

185 293 

186If you are on a managed enterprise laptop, the best long-term fix is usually to294For a managed enterprise laptop, check [MXC compatibility](#mxc-compatibility)

187get the `elevated` sandbox working with help from your IT team.295first. If MXC is unavailable or disabled, ask your IT team to provision

296`elevated`.

188 297 

189 298 

190 299 


275What to try:384What to try:

276 385 

2771. Restart Codex.3861. Restart Codex.

2782. Try the `elevated` sandbox setup again.3872. For MXC, repeat the [compatibility probe](#mxc-compatibility) and check the

2793. If that does not fix it, use the `unelevated` sandbox as a temporary388 effective network policy. For the classic `elevated` implementation, try

280 fallback.389 sandbox setup again.

3903. If a classic sandbox is needed and managed policy permits it, use

391 `unelevated` as a temporary fallback.

2814. Collect the sandbox log for review.3924. Collect the sandbox log for review.

282 393 

283 394 


296It is also helpful to include:407It is also helpful to include:

297 408 

298- a short description of what you were trying to do,409- a short description of what you were trying to do,

299- whether the `elevated` sandbox failed or the `unelevated` sandbox was used,410- the selected implementation: `mxc`, `elevated`, or `unelevated`,

300- any error message shown in the app,411- any error message shown in the app,

301- whether you saw `1385` or another Windows or PowerShell error,412- whether you saw `1385` or another Windows or PowerShell error,

413- your Windows build number,

302- and whether you are on Windows 11 or Windows 10.414- and whether you are on Windows 11 or Windows 10.

303 415 

304Do not send:416Do not send:

windows/wsl.md +5 −5

Details

5When you use WSL2, Codex runs inside the Linux environment instead of using the5When you use WSL2, Codex runs inside the Linux environment instead of using the

6native [Windows sandbox](https://learn.chatgpt.com/docs/windows/windows-sandbox). Choose WSL2 when you need Linux-native6native [Windows sandbox](https://learn.chatgpt.com/docs/windows/windows-sandbox). Choose WSL2 when you need Linux-native

7tooling, your repositories and developer workflow already live in WSL2, or7tooling, your repositories and developer workflow already live in WSL2, or

8neither native Windows sandbox mode works for your environment.8the available Windows sandbox implementations don't meet your needs.

9 9 

10WSL1 was supported through Codex `0.114`. Starting in Codex `0.115`, the Linux10WSL1 was supported through Codex `0.114`. Starting in Codex `0.115`, the Linux

11sandbox moved to `bubblewrap`, so WSL1 is no longer supported.11sandbox moved to `bubblewrap`, so WSL1 is no longer supported.


41 41 

42 This prints your distribution name.42 This prints your distribution name.

43 43 

44If you don't see "WSL: ..." in the status bar, press `Ctrl+Shift+P`, pick44If you don't see "WSL: …" in the status bar, press `Ctrl+Shift+P`, pick `WSL:

45 `WSL: Reopen Folder in WSL`, and keep your repository under `/home/...` (not45 Reopen Folder in WSL`, and keep your repository under `/home/...` (not `C:\`)

46 `C:\`) for best performance.46 for best performance.

47 47 

48If the Windows app or project picker does not show your WSL repository, type48If the Windows app or project picker does not show your WSL repository, type

49 `\\wsl$` into the file picker or Explorer, then navigate to your49 `\\wsl$` into the file picker or Explorer, then navigate to your

50 distro's home directory.50 distribution's home directory.

51 51 

52## Use Codex CLI with WSL52## Use Codex CLI with WSL

53 53