Feat/codex max models #44

cau1k · 2025-11-19T19:10:48Z

Closes #43. Adds full gpt-5.1-codex-max support with new model and reasoning effort (xhigh/extra high). Tests updated to cover the new model. User tested with pro account.

Largely written by gpt-5.1-codex-max (:

…igh), add tests

cau1k · 2025-11-20T05:41:04Z

Thought I had a ghost in the machine when trying to test out my implementation in the main opencode repo. Turns out it has this plugin already initialized in its .opencode config lol

iamhenry · 2025-11-20T17:05:44Z

looking forward to it!

numman-ali · 2025-11-20T21:37:39Z

This is very good @cau1k - going to finalise checks tonight, merge and then publish

Would you consider being a permanent collaborator? You're the first one to cover it to the same patterns/standards I go for ^_^

Copilot

Pull Request Overview

This pull request adds comprehensive support for the GPT-5.1 Codex Max model family, including the new xhigh (extra high) reasoning effort level. The changes introduce 5 new Codex Max presets (base, low, medium, high, and xhigh), expand the type system to support new reasoning options (none/xhigh for effort, off/on for summary), and update all configuration files, documentation, and tests to reflect the new capabilities.

Key Changes

Adds GPT-5.1 Codex Max model normalization and reasoning configuration with xhigh effort support (exclusive to Codex Max)
Extends type definitions to include none/xhigh reasoning effort and off/on summary options
Updates configuration files, documentation, and test scripts with 5 new Codex Max presets and expanded output limits

Reviewed Changes

Copilot reviewed 13 out of 14 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
lib/types.ts	Extended `ConfigOptions` and `ReasoningConfig` to include new reasoning effort values (`none`, `xhigh`) and summary options (`off`, `on`)
lib/request/request-transformer.ts	Added Codex Max detection, default effort of `high`, and logic to preserve `xhigh` only for Codex Max while downgrading it to `high` for other models
lib/request/helpers/model-map.ts	Added 5 new Codex Max preset mappings (base, low, medium, high, xhigh) all normalizing to `gpt-5.1-codex-max`
test/request-transformer.test.ts	Added tests for Codex Max normalization, default effort, xhigh preservation, and downgrade behavior for non-Max models; updated default effort expectations
scripts/test-all-models.sh	Added 4 Codex Max test scenarios with different reasoning levels (low, medium, high, xhigh)
config/full-opencode.json	Added 5 Codex Max model configurations with varied reasoning efforts and output limits
docs/getting-started.md	Added Codex Max configuration examples and usage notes explaining expanded output capabilities
docs/configuration.md	Documented Codex Max-specific reasoning options and behavior differences from standard Codex
docs/index.md	Updated feature list to reflect 12 GPT 5.1 variants including Codex Max
README.md	Added Codex Max configurations, usage examples, and updated model comparison table
CHANGELOG.md	Documented version 3.3.0 changes including Codex Max support and new reasoning options
AGENTS.md	Updated architecture documentation to include Codex Max in model normalization flow
config/README.md	Updated variant count and descriptions to include Codex Max presets
package-lock.json	Dependency resolution changes (unrelated to feature)

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-11-20T21:43:44Z

config/full-opencode.json

+        "gpt-5.1-codex-max": {
+          "name": "GPT 5.1 Codex Max (OAuth)",
+          "limit": {
+            "context": 272000,
+            "output": 128000
+          },
+          "options": {
+            "reasoningEffort": "high",
+            "reasoningSummary": "detailed",
+            "textVerbosity": "medium",
+            "include": [
+              "reasoning.encrypted_content"
+            ],
+            "store": false
+          }
+        },


Inconsistent output limit for gpt-5.1-codex-max. The documentation in README.md and docs/getting-started.md specifies 400000 for the base gpt-5.1-codex-max preset, indicating expanded output capacity, but this configuration file uses 128000. This should be updated to 400000 to match the documented capabilities and the other configuration files.

The correct value is 272000

OpenAIs context of 400000 comes from the combination of input and output

See here openai/codex#4728

Copilot · 2025-11-20T21:43:45Z

config/full-opencode.json

+        "gpt-5.1-codex-max-xhigh": {
+          "name": "GPT 5.1 Codex Max Extra High (OAuth)",
+          "limit": {
+            "context": 272000,
+            "output": 128000
+          },
+          "options": {
+            "reasoningEffort": "xhigh",
+            "reasoningSummary": "detailed",
+            "textVerbosity": "medium",
+            "include": [
+              "reasoning.encrypted_content"
+            ],
+            "store": false
+          }
+        },


Inconsistent output limit for gpt-5.1-codex-max-xhigh. The documentation in README.md and docs/getting-started.md specifies 400000 for the xhigh preset, indicating it should have the expanded output window, but this configuration file uses 128000. This should be updated to 400000 to match the documented capabilities and the other configuration files.

The correct value is 272000

OpenAIs context of 400000 comes from the combination of input and output

See here openai/codex#4728

CHANGELOG.md

README.md

cau1k · 2025-11-20T22:00:50Z

@numman-ali Of course, happy to help (: Feel free to reach out on discord @ caulkenstein

Good catches by copilot!

Co-authored-by: Copilot <[email protected]>

numman-ali

Config changes are required as they affect compaction behaviour in opencode

Normally, I tell my coding agent to read all comments on the PR (using gh cli command)

And it then goes ahead with rectifying everything and I just review

docs/getting-started.md

README.md

hellowodl · 2025-11-22T04:59:44Z

openai/codex#6879
openai/codex#6894

It might be worth it to look into these 2 PRs, as this is the main appeal of the new max models.

EvanNotFound · 2025-11-22T17:21:37Z

looking forward to the new max model!

ben-vargas · 2025-11-23T17:34:03Z

openai/codex#6879 openai/codex#6894

It might be worth it to look into these 2 PRs, as this is the main appeal of the new max models.

The opencode compaction logic isn't in the purview of plugins, opencode doesn't expose plugin hooks related to compaction events or opencode storage access. The logic for opencode compaction would need to be modified in opencode's source IF Dax was looking to adopt/use similar logic as codex does client side and then calling the compaction endpoint. I don't think it makes sense, better all opencode compaction logic is consistent and doesn't change behavior depending on authentication type (i.e. OAuth use shouldn't differ from using OpenAI models through zen, vercel, openrouter, etc).

cau1k · 2025-11-24T00:10:19Z

If you are eager to try out codex max in opencode, I'd recommend using https://github.com/open-hax/codex or pulling this PR:

gh repo clone numman-ali/opencode-openai-codex-auth
cd opencode-openai-codex-auth
gh pr checkout 44
Add "plugin": [ "file://path/to/opencode-openai-codex-auth" ] to your opencode config

iamhenry · 2025-11-24T02:39:00Z

If you are eager to try out codex max in opencode, I'd recommend using https://github.com/open-hax/codex or pulling this PR:

gh repo clone numman-ali/opencode-openai-codex-auth

cd opencode-openai-codex-auth

gh pr checkout 44

Add "plugin": [ "file://path/to/opencode-openai-codex-auth" ] to your opencode config

I might try this! Then I can remove the repo from my local machine?

ben-vargas · 2025-11-24T02:56:02Z

If you are eager to try out codex max in opencode, I'd recommend using https://github.com/open-hax/codex or pulling this PR:

gh repo clone numman-ali/opencode-openai-codex-auth

cd opencode-openai-codex-auth

gh pr checkout 44

Add "plugin": [ "file://path/to/opencode-openai-codex-auth" ] to your opencode config

I would actually advise the clone approach for all users whether getting this feature early or not. Not that I don't trust maintainer or contributors (I'm one), but the dependency isn't tracked, it's auto updated periodically by opencode with no notice, it opens an attack vector for npm hijacking.

You could also pin the version so you have more control as an alternative, but I like having copies of the source too just so I can look into what's going on if I want to tweak something. So I do this method (git clone) for all the opencode plugins I run.

chmistj · 2025-11-25T11:27:59Z

docs/configuration.md

 - `high` - Maximum code quality

+**GPT-5.1-Codex-Max Values:**
+- `none` - No dedicated reasoning phase


none does not seem to be supported for GPT-5.1-Codex-Max:

[openai-codex-plugin] 400 error: { "error": { "message": "Unsupported value: 'none' is not supported with the 'gpt-5.1-codex-max' model. Supported values are: 'low', 'medium', 'high', and 'xhigh'.", "type": "invalid_request_error", "param": "reasoning.effort", "code": "unsupported_value", "status": 400 } }

Surprisingly, it works with GPT-5.1, while the documented minimal errors out:

[openai-codex-plugin] 400 error: { "error": { "message": "Unsupported value: 'minimal' is not supported with the 'gpt-5.1' model. Supported values are: 'none', 'low', 'medium', and 'high'.", "type": "invalid_request_error", "param": "reasoning.effort", "code": "unsupported_value", "status": 400 } }

numman-ali · 2025-11-25T13:58:32Z

Sorry, been quite busy, merging in now and reviewing final updates before release

For everyone watching, I think this might be the last update, as OpenCode have confirmed they are working on their own plugin so only a matter of time before this becomes redundant

cau1k added 10 commits November 19, 2025 13:50

feat: update docs

1cabd2e

feat: more docs

c85dda1

feat: update model map to add gpt-5.1-codex-max

a058c21

feat: adjust types and request for codex high (high and xhigh/extra h…

90d04fb

…igh), add tests

feat: update test script

f57d8b9

feat: update changelog

eb71373

feat: npm

8bb8912

fix: add missing models

54f62d8

fix: docs

6b39290

fix: model normalization

aeb5197

numman-ali requested a review from Copilot November 20, 2025 21:37

Copilot started reviewing on behalf of numman-ali November 20, 2025 21:38 View session

numman-ali self-assigned this Nov 20, 2025

numman-ali added the enhancement New feature or request label Nov 20, 2025

Copilot finished reviewing on behalf of numman-ali November 20, 2025 21:42

Copilot AI reviewed Nov 20, 2025

View reviewed changes

cau1k and others added 2 commits November 20, 2025 17:04

Update README.md

5b5e1e9

Co-authored-by: Copilot <[email protected]>

Update CHANGELOG.md

9bdbc5f

Co-authored-by: Copilot <[email protected]>

ben-vargas mentioned this pull request Nov 20, 2025

feat: Support gpt-5.1-codex-max and xhigh reasoning #45

Closed

numman-ali requested changes Nov 20, 2025

View reviewed changes

docs: remove stray 400k references

b58f746

cau1k requested a review from numman-ali November 20, 2025 22:55

chmistj reviewed Nov 25, 2025

View reviewed changes

numman-ali merged commit c5afe9a into numman-ali:main Nov 25, 2025

Feat/codex max models #44

Feat/codex max models #44

Uh oh!

Conversation

cau1k commented Nov 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cau1k commented Nov 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

iamhenry commented Nov 20, 2025

Uh oh!

numman-ali commented Nov 20, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Key Changes

Reviewed Changes

Uh oh!

Copilot AI Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

numman-ali Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

numman-ali Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

cau1k commented Nov 20, 2025

Uh oh!

numman-ali left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hellowodl commented Nov 22, 2025

Uh oh!

EvanNotFound commented Nov 22, 2025

Uh oh!

ben-vargas commented Nov 23, 2025

Uh oh!

cau1k commented Nov 24, 2025

Uh oh!

iamhenry commented Nov 24, 2025

Uh oh!

ben-vargas commented Nov 24, 2025

Uh oh!

chmistj Nov 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

chmistj Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

numman-ali commented Nov 25, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

cau1k commented Nov 19, 2025 •

edited

Loading

cau1k commented Nov 20, 2025 •

edited

Loading

chmistj Nov 25, 2025 •

edited

Loading