fix: max reasoning effort in openai

[TejasGhatte]

Summary

Adds support for context and mode fields in the ResponsesParametersReasoning schema, and extends the reasoning effort normalization logic to treat gpt-5.6 as a model that natively supports "max" effort (and "xhigh"), bypassing the downgrade path applied to earlier GPT-5.x models.

Changes

• Added Context and Mode fields to ResponsesParametersReasoning, allowing callers to specify which reasoning items are rendered back to the model on later turns (context) and the reasoning execution mode (mode).
• Propagated Context and Mode through MarshalJSON on OpenAIResponsesRequest so they are included in the serialized output alongside the existing Effort, GenerateSummary, and Summary fields.
• Added gpt-5.6 to supportsXHighReasoningEffort and supportsMaxReasoningEffort, so that "max" and "xhigh" efforts are passed through without downgrade for that model family.
• Added TestNormalizeOpenAIReasoningEffort to cover the effort normalization logic across multiple model prefixes and effort values.
• Extended the existing MarshalJSON test to assert that reasoning.context and reasoning.mode are preserved in the JSON output.

Type of change

• Bug fix
• Feature
• Refactor
• Documentation
• Chore/CI

Affected areas

• Core (Go)
• Transports (HTTP)
• Providers/Integrations
• Plugins
• UI (React)
• Docs

How to test

go test ./core/providers/openai/... ./core/schemas/...

• Verify TestNormalizeOpenAIReasoningEffort passes and covers gpt-5.6 keeping "max" and "xhigh" without downgrade.
• Verify the MarshalJSON test case "reasoning context and mode are preserved while max_tokens is dropped" passes, confirming context and mode appear in the serialized JSON while max_tokens is omitted.

Breaking changes

• Yes
• No

Related issues

Security considerations

No security implications. These are additive schema fields passed through to the OpenAI Responses API.

Checklist

• I read docs/contributing/README.md and followed the guidelines
• I added/updated tests where appropriate
• I updated documentation where needed
• I verified builds succeed (Go and UI)
• I verified the CI pipeline passes locally if applicable

[coderabbitai]

📝 Walkthrough

Walkthrough

OpenAI reasoning parameters now support optional context and mode fields, preserve them during response request marshaling, omit max_tokens, and recognize additional models for xhigh and max reasoning-effort normalization. Tests cover serialization and normalization cases.

Changes

OpenAI reasoning configuration

Layer / File(s)	Summary
Reasoning fields and serialization core/schemas/responses.go, core/providers/openai/types.go, core/providers/openai/responses_marshal_test.go	Reasoning parameters add optional context and mode; OpenAI marshaling preserves these fields while omitting max_tokens, with corresponding JSON assertions.
Reasoning-effort model classification core/providers/openai/utils.go, core/providers/openai/responses_marshal_test.go	Reasoning-effort support includes gpt-5.5 and gpt-5.6, and normalization tests cover model-specific effort mappings.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• maximhq/bifrost#4467: Updates OpenAI reasoning-effort model handling and tests for preserving max.

Suggested reviewers: akshaydeo, danpiths

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Linked Issues check	⚠️ Warning	The PR changes reasoning-effort handling, not the requested Files API support in issue #123.	Implement the Files API support requested in #123 or retarget this PR to the correct linked issue.
Out of Scope Changes check	⚠️ Warning	All code changes are unrelated to the linked Files API objective and appear out of scope for #123.	Remove the reasoning-effort changes or split them into a separate PR tied to the correct issue.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Title check	✅ Passed	The title accurately summarizes the main change: OpenAI reasoning-effort handling, including max effort support.
Description check	✅ Passed	The description covers the required sections, with summary, changes, testing, and status details filled in adequately.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch 07-13-fix_max_reasoning_effort_in_openai

---

_{Comment @coderabbitai help to get the list of available commands.}

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

tejas ghatte seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[TejasGhatte]

• fix: max reasoning effort in openai #5130 👈 (View in Graphite)
• dev

This stack of pull requests is managed by Graphite. Learn more about stacking.

[greptile-apps]

Confidence Score: 4/5

The changed OpenAI Responses serialization path needs fixes before merging.

• Compatible-provider Responses requests can now receive new reasoning keys they may not accept.
• Context-only or mode-only reasoning requests can serialize effort as null.
• The GPT-5.6 effort normalization change is small and covered by direct tests.

core/providers/openai/types.go

Important Files Changed

Filename	Overview
core/providers/openai/types.go	The Responses marshaler now forwards reasoning.context and reasoning.mode, which can affect compatible-provider payloads and context-only reasoning requests.
core/providers/openai/utils.go	Reasoning effort normalization now keeps max and xhigh for GPT-5.6 models.
core/schemas/responses.go	The shared Responses reasoning schema now includes optional context and mode fields and documents wider effort values.
core/providers/openai/responses_marshal_test.go	Tests now cover context/mode preservation and GPT-5.6 reasoning effort normalization.

_{Reviews (1): Last reviewed commit: "fix: max reasoning effort in openai" | Re-trigger Greptile}

[Pratham-Mishra04]

Merge activity

• Jul 13, 9:03 AM UTC: A user started a stack merge that includes this pull request via Graphite.
• Jul 13, 9:04 AM UTC: @Pratham-Mishra04 merged this pull request with Graphite.