GPT5 OpenAI Fix by hannesrudolph · Pull Request #6864 · RooCodeInc/Roo-Code

hannesrudolph · 2025-08-09T01:00:02Z

Summary

This PR implements comprehensive GPT-5 support including the Responses API, conversation continuity, and proper token management.

Major Changes

1. Complete GPT-5 Responses API Implementation

Full streaming support via OpenAI SDK with fallback to fetch-based SSE
Conversation continuity using previous_response_id for efficient multi-turn conversations
Automatic reasoning summaries with summary: "auto"
Support for all reasoning effort levels including new "minimal" level
Comprehensive event handling for 40+ different GPT-5 streaming event types
Response ID tracking and persistence

2. Token Management

Added explicit max_output_tokens parameter to prevent defaulting to excessive limits (e.g., 120k)
Uses Roo's calculated reserved output tokens via model.maxTokens
Applies 20% clamping rule for output tokens

3. Task Persistence Layer

Stores GPT-5 metadata per conversation turn:
- previous_response_id for conversation continuity
- Instructions used for each turn
- Reasoning summaries
Enables efficient context management across long conversations

4. UI and Settings Updates

Updated thinking budget interface
Modified API options display
Added GPT-5-specific settings

5. Comprehensive Test Coverage

950+ lines of new tests
Complete coverage of GPT-5 streaming events
Tests for both SDK and fallback SSE paths
Validation of max_output_tokens inclusion

6. Internationalization

Added GPT-5 related translations across all 18 supported locales

Files Changed

Core Implementation: src/api/providers/openai-native.ts (1000+ lines)
Tests: src/api/providers/__tests__/openai-native.spec.ts (950+ lines)
Task Persistence: src/core/task/Task.ts
Type Definitions: packages/types/src/ files
UI Components: webview-ui/src/components/settings/
Translations: All locale files (18 languages)

Testing

✅ All 40 tests in openai-native.spec.ts passing
✅ Verified GPT-5 streaming with all event types
✅ Confirmed max_output_tokens properly included
✅ Tested conversation continuity with previous_response_id
✅ Validated fallback from SDK to SSE

Impact

This is a significant feature addition that enables full GPT-5 support with proper token management, conversation continuity, and comprehensive error handling.

- Added max_output_tokens parameter to GPT-5 request body using model.maxTokens - This prevents GPT-5 from defaulting to very large token limits (e.g., 120k) - Updated tests to expect max_output_tokens in GPT-5 request bodies - Fixed test for handling unhandled stream events by properly mocking SDK fallback

Copilot

Pull Request Overview

This PR adds a new minimal reasoning effort level for GPT-5 models and updates all internationalization files to include translations for this new option. The minimal effort provides the fastest time-to-first-token response from GPT-5 models.

Adds minimal reasoning effort support for GPT-5 models with internationalization
Updates UI components to conditionally show the minimal option for GPT-5 models
Modifies reasoning transformation logic to handle the new minimal effort level

Reviewed Changes

Copilot reviewed 31 out of 31 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
webview-ui/src/i18n/locales/*/settings.json	Adds "minimal (fastest)" translation entries across 16 locales
webview-ui/src/components/settings/ThinkingBudget.tsx	Adds GPT-5 model detection and conditional minimal reasoning effort option
webview-ui/src/components/settings/ApiOptions.tsx	Clears reasoning effort when switching models and shows verbosity only for GPT-5
src/shared/api.ts	Adds enableGpt5ReasoningSummary option to ApiHandlerOptions
src/core/task/Task.ts	Implements GPT-5 conversation continuity with previous_response_id metadata
src/api/transform/reasoning.ts	Updates reasoning transformations to exclude minimal effort from API calls
src/api/transform/model-params.ts	Updates type definitions for ReasoningEffortWithMinimal
src/api/providers/requesty.ts	Filters out minimal effort from API requests
src/api/providers/openai.ts	Updates type casting for reasoning effort parameters
src/api/providers/openai-native.ts	Major refactor for GPT-5 Responses API support with conversation continuity
src/api/providers/tests/openai-native.spec.ts	Comprehensive test updates for GPT-5 Responses API behavior
src/api/index.ts	Adds previousResponseId to metadata interface
packages/types/src/providers/openai.ts	Adds GPT-5 model default reasoning effort and temperature
packages/types/src/provider-settings.ts	Defines ReasoningEffortWithMinimal type and schema
packages/types/src/message.ts	Adds GPT-5 metadata schema for conversation continuity

ghost

Thank you for your contribution! I've reviewed the changes and found that the PR successfully addresses the stated problem of adding explicit max_output_tokens for GPT-5 requests. The implementation goes beyond the initial scope by also adding minimal reasoning effort support and conversation continuity features.

Positive observations:

Excellent test coverage with 40+ tests including comprehensive GPT-5 functionality
Good separation of concerns with dedicated methods for GPT-5 handling
Proper implementation of the max_output_tokens parameter as intended
Comprehensive handling of various GPT-5 response formats

I've left some suggestions inline that could improve code maintainability and reduce duplication.

…an and Dutch locales

- Renamed metadata field from 'previous_response_id' to 'response_id' for clarity - Fixed logic to correctly use the response_id from the previous message as previous_response_id for the next request - This resolves the 'Previous response with id not found' errors that occurred after multiple turns in the same session

- Automatically retry without previous_response_id when it's not found (400 error) - Clear stored lastResponseId to prevent reusing stale IDs - Handle errors in both SDK and SSE fallback paths - Log warnings when retrying to help with debugging

- Add promise-based synchronization for response ID persistence - Wait for pending response ID from previous request before using it - Resolve promise when response ID is received or cleared - Add 100ms timeout to avoid blocking too long on ID resolution - Properly clean up resolver on errors to prevent memory leaks This fixes the race condition where fast nano model responses could cause the next request to be initiated before the response ID was fully persisted.

hannesrudolph

Good Job ROO!

- Extract usage normalization helper to reduce duplication - Suppress conversation continuity for first message (but respect explicit metadata) - Deduplicate response ID resolver logic - Remove dead enableGpt5ReasoningSummary option references - DRY up GPT-5 event/usage handling with normalizeGpt5Usage helper - Centralize default GPT-5 reasoning effort using model info - Fix Indonesian locale minimal string misplacement - Add clarifying comments for Developer prefix usage - Add TODO for future verbosity UI capability gating - Fix failing test in reasoning.spec.ts

…ndard GPT-5 SSE event types to shared processor to reduce duplication\n- Add JSDoc for response ID accessors\n- Standardize key error messages for GPT-5 Responses API fallback\n- Extract persistGpt5Metadata() in Task to simplify metadata writes\n- Add malformed JSON SSE parsing test\n

…tOpenAI incl. cache); enforce 'skip once' continuity via suppressPreviousResponseId; dedupe responseId resolver on SSE 400; feat: gate reasoning.summary by enableGpt5ReasoningSummary; centralize default reasoning effort; types/ui: add ModelInfo.supportsVerbosity and gate Verbosity UI by capability; refactor: avoid duplicate usage emission in SSE done/completed

…and expected behavior

…d align enableGpt5ReasoningSummary default docs

…n completePrompt

daniel-lxs

LGTM

Removes unused translation strings introduced in PR #6864 that are no longer needed after PR #6921 refactoring. Keys removed: includeMaxOutputTokens, includeMaxOutputTokensDescription, maxOutputTokensLabel, maxTokensGenerateDescription

This reverts commit cda67a8.

- Fix manual condensing bug by setting skipPrevResponseIdOnce flag in Task.ts - Revert to string-based format for full conversations after condensing - Add proper image support with structured format when using previous_response_id - Update test suite to match new implementation where all models use Responses API - Handle 400 errors gracefully when previous_response_id is not found Fixes issues introduced in PR #6864

CopilotAI review requested due to automatic review settings August 9, 2025 01:00

hannesrudolph requested review from cte, jr and mrubens as code owners August 9, 2025 01:00

github-project-automationBot added this to Roo Code Roadmap and Roo Code Roadmap Aug 9, 2025

github-project-automationBot moved this to Triage in Roo Code Roadmap Aug 9, 2025

github-project-automationBot moved this to New in Roo Code Roadmap Aug 9, 2025

dosubotBot added size:XXL This PR changes 1000+ lines, ignoring generated files. bug Something isn't working labels Aug 9, 2025

CopilotAI reviewed Aug 9, 2025
View reviewed changes

Comment threadwebview-ui/src/i18n/locales/id/settings.json Outdated
Comment threadwebview-ui/src/components/settings/ThinkingBudget.tsx
Comment threadsrc/api/providers/openai-native.ts
Comment threadsrc/api/providers/openai-native.ts
Comment threadsrc/api/providers/__tests__/openai-native.spec.ts

hannesrudolph changed the title ~~fix: add explicit max_output_tokens for GPT-5 Responses API~~GPT5 OpenAI FixAug 9, 2025

hannesrudolph added the Issue/PR - Triage New issue. Needs quick review to confirm validity and assign labels. label Aug 9, 2025

ghost reviewed Aug 9, 2025
View reviewed changes

hannesrudolph added 4 commits August 8, 2025 19:06

fix: add missing translations for reasoningEffort.minimal in Indonesi…
f749d35
…an and Dutch locales

hannesrudolph commented Aug 9, 2025
View reviewed changes

hannesrudolph added 6 commits August 8, 2025 19:57

fix(gpt5): default enableGpt5ReasoningSummary=true to preserve tests …
614dc44
…and expected behavior

fix(gpt5): canonicalize GPT-5 metadata key to previous_response_id an…
64a2a03
…d align enableGpt5ReasoningSummary default docs

fix(openai-native): remove review artifact comments and guard GPT-5 i…
17e7e40
…n completePrompt

daniel-lxs moved this from Triage to PR [Needs Review] in Roo Code Roadmap Aug 9, 2025

daniel-lxs approved these changes Aug 9, 2025
View reviewed changes

dosubotBot added the lgtm This PR has been approved by a maintainer label Aug 9, 2025

hannesrudolph removed the Issue/PR - Triage New issue. Needs quick review to confirm validity and assign labels. label Aug 9, 2025

hannesrudolph added the PR - Needs Review label Aug 9, 2025

mrubens approved these changes Aug 9, 2025
View reviewed changes

mrubens merged commit cda67a8 into mainAug 9, 2025
16 checks passed

mrubens deleted the fix/gpt5-max-output-tokens branch August 9, 2025 18:52

github-project-automationBot moved this from PR [Needs Review] to Done in Roo Code Roadmap Aug 9, 2025

github-project-automationBot moved this from New to Done in Roo Code Roadmap Aug 9, 2025

hannesrudolph mentioned this pull request Aug 11, 2025
chore: remove unused i18n keys for include max output tokens #6935
Closed
7 tasks

daniel-lxs added a commit that referenced this pull request Aug 11, 2025
Revert "GPT5 OpenAI Fix (#6864)"
6f86819
This reverts commit cda67a8.

daniel-lxs mentioned this pull request Aug 11, 2025
Revert "feat: OpenAI provider/types/UI updates; provider state persistence" #6949
Closed

daniel-lxs mentioned this pull request Aug 13, 2025
Fix GPT-5 Responses API issues with condensing and image support #7067
Merged

Navigation Menu

Uh oh!

GPT5 OpenAI Fix - #6864

GPT5 OpenAI Fix#6864
mrubens merged 11 commits into
mainfrom
fix/gpt5-max-output-tokens

hannesrudolph commented Aug 9, 2025•
edited
Loading

Uh oh!

CopilotAI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ghost left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hannesrudolph left a comment •
edited
Loading

Uh oh!

daniel-lxs left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Conversation

hannesrudolph commented Aug 9, 2025• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Major Changes

1. Complete GPT-5 Responses API Implementation

2. Token Management

3. Task Persistence Layer

4. UI and Settings Updates

5. Comprehensive Test Coverage

6. Internationalization

Files Changed

Testing

Impact

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ghost left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hannesrudolph left a comment • 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!

daniel-lxs left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

hannesrudolph commented Aug 9, 2025•
edited
Loading

hannesrudolph left a comment •
edited
Loading