Add GPT-5.1 models and clean up reasoning effort logic #9252

hannesrudolph · 2025-11-14T00:03:33Z

Reasoning effort is now capability‑driven with disable/none/minimal; removed GPT‑5 special‑casing

Description

Why
- Eliminate brittle GPT‑5 “minimal” band‑aids.
- Give models/providers precise control over which reasoning efforts are exposed and how they’re sent.
- Clearly distinguish “disable” (omit param) vs “none” (send param=none).
What changed for users
- The Reasoning Effort dropdown now reflects the model’s capability:
  - If the model advertises boolean support, you’ll see Low/Medium/High.
  - If the model advertises an explicit list, you’ll see exactly those values (any mix of Disable, None, Minimal, Low, Medium, High).
- Selecting:
  - Disable: turns reasoning off and sends no reasoning parameter.
  - None: keeps reasoning enabled but explicitly sends “none”.
  - Minimal/Low/Medium/High: sends the selected effort.
- Models that require reasoning effort won’t show a synthetic “None”; they use the model’s default if unset.
Why it matters
- Removes hidden provider‑specific behavior.
- Enables explicit opt‑out (Disable) versus explicit “no reasoning” configuration (None).
- Reduces user confusion and mismatches between UI and actual provider capabilities.
Backward compatibility
- Models with supportsReasoningEffort=true still show Low/Medium/High.
- Existing settings continue to work; additional values are honored when supported.
Additional notes
- OpenAI Native “minimal” no longer suppressed; reasoning fields and encrypted content only included when an effort is selected.
- Temporary OpenAI Native request body logs removed.

Important

This PR refactors reasoning effort handling to be capability-driven, removing GPT-5 special-casing and updating models, API handlers, and UI components accordingly.

Behavior:
- Introduces capability-driven reasoning effort handling with options: disable, none, minimal, low, medium, high.
- Removes special-casing for GPT-5 models, allowing explicit control over reasoning efforts.
- Updates SimpleThinkingBudget and ThinkingBudget components to reflect new reasoning effort logic.
Models:
- Updates model.ts to include reasoningEffortsExtended and reasoningEffortSettingValues.
- Modifies openai.ts to define new GPT-5.1 models with updated reasoning capabilities.
API Handlers:
- Refactors OpenAiNativeHandler and RooHandler to utilize new reasoning effort logic.
- Adjusts getModelParams and getOpenAiReasoning functions to handle extended reasoning efforts.
Tests:
- Updates and adds tests in openai-native.spec.ts, roo.spec.ts, and model-params.spec.ts to cover new reasoning effort scenarios.
- Ensures UI components handle reasoning effort changes correctly in SimpleThinkingBudget.spec.tsx.

^{This description was created by}^{for f1aa7f7. You can customize this summary. It will automatically update as commits are pushed.}

… GPT-5 minimal special-casing; document UI semantics; remove temporary logs

roomote · 2025-11-14T00:03:53Z

Rooviewer See task on Roo Cloud

Re-reviewed changes since last commit. Three issues resolved, one remaining.

Logic bug in getRooReasoning: The function treats reasoningEffort === "disable" the same as enableReasoningEffort === false, both returning { enabled: false }. However, "disable" should return undefined to omit the reasoning parameter entirely, as expected by the test at line 889.
Dead code: The supportsReasoningNone field is added to the ModelInfo schema but never used anywhere in the codebase.
State inconsistency in SimpleThinkingBudget: When "disable" is selected, the UI sets both enableReasoningEffort = false and reasoningEffort = "disable". However, getRooReasoning checks enableReasoningEffort first (line 56), returning { enabled: false } instead of the intended undefined, contradicting the documented goal to "omit any reasoning parameter/body sections".

Previous reviews

9345d77: Review #1

f1aa7f7: Review #2

3dfd393: Review #3

1e068a7: Review #4

1e068a7: Review #5

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}

src/api/transform/reasoning.ts

packages/types/src/model.ts

roomote · 2025-11-14T00:22:21Z

Oroocle See task on Roo Cloud

Posted 2 inline review comments addressing a logic bug and an unused schema field. Waiting on updates from the author.

getRooReasoning: map "disable" to omission (undefined) and keep { enabled: false } only when enableReasoningEffort === false. Adjust getRooReasoning() to omit the field for "disable".
Remove or document unused supportsReasoningNone in modelInfoSchema.

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}

src/api/transform/reasoning.ts

packages/types/src/model.ts

…ngEffort when selecting 'disable'

packages/types/src/model.ts

Co-authored-by: roomote[bot] <219738659+roomote[bot]@users.noreply.github.com>

webview-ui/src/components/settings/SimpleThinkingBudget.tsx

Co-authored-by: roomote[bot] <219738659+roomote[bot]@users.noreply.github.com>

Reasoning effort: capability-driven; add disable/none/minimal; remove…

1e068a7

… GPT-5 minimal special-casing; document UI semantics; remove temporary logs

hannesrudolph requested review from cte, jr and mrubens as code owners November 14, 2025 00:03

github-project-automation bot added this to Roo Code Roadmap and Roo Code Roadmap Nov 14, 2025

github-project-automation bot moved this to Triage in Roo Code Roadmap Nov 14, 2025

github-project-automation bot moved this to New in Roo Code Roadmap Nov 14, 2025

dosubot bot added size:XL This PR changes 500-999 lines, ignoring generated files. Enhancement New feature or request labels Nov 14, 2025

daniel-lxs approved these changes Nov 14, 2025

View reviewed changes

dosubot bot added the lgtm This PR has been approved by a maintainer label Nov 14, 2025

daniel-lxs moved this from Triage to PR [Needs Review] in Roo Code Roadmap Nov 14, 2025

hannesrudolph added the PR - Needs Review label Nov 14, 2025