fix: scope gpt-5.4-mini chat reasoning_effort fallback by steipete · Pull Request #76727 · openclaw/openclaw

steipete · 2026-05-03T14:26:28Z

Summary

Fixes #76176 by handling a narrow OpenAI API incompatibility for gpt-5.4-mini on /v1/chat/completions.

gpt-5.4-mini does support reasoning effort. The failing case is specifically Chat Completions requests that include both function tools and reasoning_effort. OpenAI rejects that combination, which made Telegram-routed fallback runs fail before producing a reply.

Live Verification

Verified against the OpenAI API on 2026-05-03:

Request	Result
Chat Completions + function tools, no `reasoning_effort`	HTTP 200
Chat Completions + function tools + `reasoning_effort`	HTTP 400, `param=reasoning_effort`
Chat Completions + `reasoning_effort`, no tools	HTTP 200
Responses + `reasoning.effort`	HTTP 200

The OpenAI error for the rejected case says function tools with reasoning_effort are not supported for gpt-5.4-mini in /v1/chat/completions and recommends using /v1/responses.

Root Cause

OpenClaw treated model-level reasoning support as sufficient to always emit Chat Completions reasoning_effort. That is correct for tool-free gpt-5.4-mini Chat Completions and for Responses, but not for Chat Completions payloads with function tools.

What Changed

Keeps gpt-5.4-mini reasoning-effort support in the model metadata.
Keeps reasoning_effort on tool-free gpt-5.4-mini Chat Completions payloads.
Keeps reasoning.effort on gpt-5.4-mini Responses payloads.
Omits reasoning_effort only when a native OpenAI Chat Completions payload for gpt-5.4-mini includes function tools.
Adds regression coverage for both sides of the contract: tool payloads omit the parameter; tool-free payloads keep it.

Scope

OpenAI Chat Completions payload compatibility only. No Telegram transport behavior, model fallback policy, provider routing, auth, credentials, or Responses behavior changes.

Test Plan

Live OpenAI API matrix above.
pnpm exec oxfmt --write --threads=1 src/agents/openai-reasoning-effort.ts src/agents/openai-reasoning-effort.test.ts src/agents/openai-transport-stream.ts src/agents/openai-transport-stream.test.ts CHANGELOG.md
pnpm test src/agents/openai-reasoning-effort.test.ts src/agents/openai-transport-stream.test.ts -- --reporter=verbose
GitHub PR CI: green on ea3915308c3c12bb6e71cdb5f13471ec0eb5ba53

Replacement

Replaces the OpenAI-only portion of closed bundled PR #76326.

clawsweeper · 2026-05-03T14:27:23Z

Codex review: needs maintainer review before merge.

Summary
The PR adds gpt-5.4-mini detection, omits Chat Completions reasoning_effort only for tool payloads, and adds focused tests plus a changelog entry.

Reproducibility: yes. source-level reproduction is high confidence: current main can build a gpt-5.4-mini Chat Completions payload with tools and reasoning_effort, matching the OpenAI 400 reported in #76176 and the PR's live matrix.

Next step before merge
No ClawSweeper repair lane is needed; the prior review blocker appears addressed and the PR should proceed through normal maintainer and CI review.

Security
Cleared: The diff only changes TypeScript OpenAI payload compatibility logic, focused tests, and changelog text; it does not touch dependencies, workflows, secrets, or package execution surfaces.

Review details

Best possible solution:

Land the scoped transport-level omission after required CI and maintainer review, keeping reasoning metadata available for Responses and tool-free Chat Completions.

Do we have a high-confidence way to reproduce the issue?

Yes, source-level reproduction is high confidence: current main can build a gpt-5.4-mini Chat Completions payload with tools and reasoning_effort, matching the OpenAI 400 reported in #76176 and the PR's live matrix.

Is this the best way to solve the issue?

Yes, the latest diff is the narrow maintainable fix: suppress only the rejected Chat Completions + function-tools combination while preserving Responses and tool-free Chat reasoning behavior.

What I checked:

Current main can emit the rejected parameter: Current main resolves gpt-5.4-mini through the GPT-5.2+ reasoning-effort branch and buildOpenAICompletionsParams assigns reasoning_effort when the model supports reasoning and completions compat allows it. (src/agents/openai-reasoning-effort.ts:83, 935f078a89f0)
Latest PR scopes the fix to tool payloads: The final diff adds isOpenAIGpt54MiniModel and suppresses reasoning_effort only when params.tools is a non-empty array, leaving tool-free Chat Completions and Responses resolver metadata intact. (src/agents/openai-transport-stream.ts:1900, ea3915308c3c)
Regression coverage matches the behavior boundary: The PR adds payload tests for gpt-5.4-mini with tools omitting reasoning_effort and without tools retaining it, plus resolver tests preserving Chat and Responses reasoning metadata. (src/agents/openai-transport-stream.test.ts:2050, ea3915308c3c)
Related report and prior PR context: The PR explicitly targets the gpt-5.4-mini Telegram/OpenAI failure from [Bug]: Telegram agent hangs with openai/gpt-5.4-mini; chat-completions fallback sends unsupported reasoning_effort #76176 and replaces the OpenAI-only portion of closed PR fix: exclude gpt-5.4-mini from reasoning_effort in /v1/chat/completions #76326, preserving the useful prior investigation while narrowing the implementation.
Protected-label handling: The provided GitHub context shows this PR carries the maintainer label, so it should remain open for explicit maintainer/CI handling rather than cleanup closure.
Feature history: Blame on the current main resolver and OpenAI transport payload builder points to commit 399d7f6 as the recent source-history owner for this surface. (src/agents/openai-transport-stream.ts:1894, 399d7f61783f)

Likely related people:

steipete: Current-main blame for the OpenAI reasoning resolver, completions payload builder, Responses reasoning path, and transport API registry points to Peter Steinberger's recent agents commit; he also authored the scoped follow-up PR. (role: recent maintainer and likely follow-up owner; confidence: high; commits: 399d7f61783f; files: src/agents/openai-reasoning-effort.ts, src/agents/openai-transport-stream.ts, src/agents/provider-transport-stream.ts)

Remaining risk / open question:

This read-only review did not execute the targeted Vitest/format checks; merge should still wait for PR CI or maintainer-run targeted validation.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 935f078a89f0.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c1deaface2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-05-03T14:50:21Z

+  if (api === "openai-responses" || api === "openai-codex-responses") {
+    if (/^gpt-5\.4-mini(?:-|$)/u.test(id)) {
+      return GPT_52_REASONING_EFFORTS;
+    }
+  } else if (/^gpt-5\.4-mini(?:-|$)/u.test(id)) {
+    return [];


Restrict gpt-5.4-mini suppression to chat completions

The new else if branch returns [] for gpt-5.4-mini on every API value except openai-responses/openai-codex-responses, which unintentionally disables reasoning for other Responses surfaces (notably azure-openai-responses used by the Responses transport path) that previously inherited GPT_52_REASONING_EFFORTS. This means Azure/OpenAI Responses requests can now silently lose reasoning effort after this commit; the suppression should be scoped to chat-completions APIs only (for example openai-completions) rather than all non-responses API strings.

Useful? React with 👍 / 👎.

steipete · 2026-05-03T15:02:45Z

Landed via squash merge onto main.

PR head: ea39153
Landed commit: 11e05e8
Gate: PR CI green on the head SHA; targeted local pnpm test src/agents/openai-reasoning-effort.test.ts src/agents/openai-transport-stream.test.ts -- --reporter=verbose

Thanks @ThisIsAdilah and @chinar-amrutkar.

@ThisIsAdilah

Fixes openclaw#76176. OpenAI live verification showed `gpt-5.4-mini` supports reasoning effort generally, but rejects `/v1/chat/completions` payloads that combine function tools with `reasoning_effort`. This keeps reasoning effort for tool-free Chat Completions and Responses, and omits it only for the rejected Chat Completions + function tools combination. Validation: - Live OpenAI API matrix on 2026-05-03 - pnpm test src/agents/openai-reasoning-effort.test.ts src/agents/openai-transport-stream.test.ts -- --reporter=verbose - GitHub PR CI green on ea39153 Thanks @ThisIsAdilah and @chinar-amrutkar.

@ThisIsAdilah

Fixes openclaw#76176. OpenAI live verification showed `gpt-5.4-mini` supports reasoning effort generally, but rejects `/v1/chat/completions` payloads that combine function tools with `reasoning_effort`. This keeps reasoning effort for tool-free Chat Completions and Responses, and omits it only for the rejected Chat Completions + function tools combination. Validation: - Live OpenAI API matrix on 2026-05-03 - pnpm test src/agents/openai-reasoning-effort.test.ts src/agents/openai-transport-stream.test.ts -- --reporter=verbose - GitHub PR CI green on ea39153 Thanks @ThisIsAdilah and @chinar-amrutkar.

@ThisIsAdilah

Fixes openclaw#76176. OpenAI live verification showed `gpt-5.4-mini` supports reasoning effort generally, but rejects `/v1/chat/completions` payloads that combine function tools with `reasoning_effort`. This keeps reasoning effort for tool-free Chat Completions and Responses, and omits it only for the rejected Chat Completions + function tools combination. Validation: - Live OpenAI API matrix on 2026-05-03 - pnpm test src/agents/openai-reasoning-effort.test.ts src/agents/openai-transport-stream.test.ts -- --reporter=verbose - GitHub PR CI green on ea39153 Thanks @ThisIsAdilah and @chinar-amrutkar.

@ThisIsAdilah

Fixes openclaw#76176. OpenAI live verification showed `gpt-5.4-mini` supports reasoning effort generally, but rejects `/v1/chat/completions` payloads that combine function tools with `reasoning_effort`. This keeps reasoning effort for tool-free Chat Completions and Responses, and omits it only for the rejected Chat Completions + function tools combination. Validation: - Live OpenAI API matrix on 2026-05-03 - pnpm test src/agents/openai-reasoning-effort.test.ts src/agents/openai-transport-stream.test.ts -- --reporter=verbose - GitHub PR CI green on ea39153 Thanks @ThisIsAdilah and @chinar-amrutkar.

@ThisIsAdilah

Fixes openclaw#76176. OpenAI live verification showed `gpt-5.4-mini` supports reasoning effort generally, but rejects `/v1/chat/completions` payloads that combine function tools with `reasoning_effort`. This keeps reasoning effort for tool-free Chat Completions and Responses, and omits it only for the rejected Chat Completions + function tools combination. Validation: - Live OpenAI API matrix on 2026-05-03 - pnpm test src/agents/openai-reasoning-effort.test.ts src/agents/openai-transport-stream.test.ts -- --reporter=verbose - GitHub PR CI green on ea39153 Thanks @ThisIsAdilah and @chinar-amrutkar.

openclaw-barnacle Bot added agents Agent runtime and tooling size: S maintainer Maintainer-authored PR labels May 3, 2026

steipete mentioned this pull request May 3, 2026

fix: exclude gpt-5.4-mini from reasoning_effort in /v1/chat/completions #76326

Closed

2 tasks

fix(agents): omit gpt-5.4-mini chat reasoning effort

c1deafa

steipete force-pushed the fix/76176-gpt54-mini-chat-reasoning branch from 2cd1c86 to c1deafa Compare May 3, 2026 14:48

chatgpt-codex-connector Bot reviewed May 3, 2026

View reviewed changes

fix(agents): scope gpt-5.4-mini chat reasoning fallback

ea39153

steipete changed the title ~~fix: omit gpt-5.4-mini chat reasoning_effort~~ fix: scope gpt-5.4-mini chat reasoning_effort fallback May 3, 2026

steipete merged commit 11e05e8 into main May 3, 2026
122 of 124 checks passed

steipete deleted the fix/76176-gpt54-mini-chat-reasoning branch May 3, 2026 15:02

martingarramon mentioned this pull request May 3, 2026

test(agents): cover isOpenAIGpt54MiniModel helper variants #76809

Closed

github-actions Bot mentioned this pull request May 3, 2026

📡 Upstream Digest — 2026-05-03 16:39 UTC curtismercier/openclaw-mods#758

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix: scope gpt-5.4-mini chat reasoning_effort fallback#76727

fix: scope gpt-5.4-mini chat reasoning_effort fallback#76727
steipete merged 2 commits into
mainfrom
fix/76176-gpt54-mini-chat-reasoning

steipete commented May 3, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented May 3, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot May 3, 2026

Uh oh!

Uh oh!

steipete commented May 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Uh oh!

Conversation

steipete commented May 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Live Verification

Root Cause

What Changed

Scope

Test Plan

Replacement

Uh oh!

clawsweeper Bot commented May 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot May 3, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

steipete commented May 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

steipete commented May 3, 2026 •

edited

Loading

clawsweeper Bot commented May 3, 2026 •

edited

Loading