update gpt-5 models

[j4ys0n]

No description provided.

[j4ys0n]

Automated review 🤖

Summary of Changes
This PR adds comprehensive support for OpenAI’s GPT-5 family of models in the SDK, including model-specific configuration for reasoning effort, verbosity, and sampling behavior. Version is bumped to 1.12.0. A new gpt5.support.ts file defines per-model capabilities (e.g., chatCompletionsSupported, allowedReasoningEfforts, supportsSampling), and the OpenAI mapper enforces these rules—dropping unsupported parameters (e.g., temperature, top_p) or throwing UnsupportedFeatureError for responses-only models like gpt-5-pro. New reasoningEffort and verbosity fields are added to GenerateParams, and unit tests cover key GPT-5 behaviors.

Key Changes & Positives

• ✅ Introduces robust, extensible model capability table (gpt5.support.ts) with exact per-model behavior definitions (e.g., gpt-5.2 supports xhigh, gpt-5-chat-latest allows sampling) 🟢
• ✅ Adds safe fallback for unknown future GPT-5 models: preserves max_tokens, drops all GPT-5-specific fields 🟢
• ✅ Enforces model constraints via UnsupportedFeatureError for non-chat-completions models (e.g., gpt-5-pro) 🟢
• ✅ Adds reasoningEffort and verbosity to public GenerateParams interface with clear JSDoc 🟢

Potential Issues & Recommendations

1. Issue / Risk: The getGpt5Support function uses model.startsWith(${key}-) for prefix matching, which may incorrectly match unrelated models (e.g., gpt-5.2-foo matches gpt-5.2, but gpt-5.20 would also match gpt-5.2).
   Impact: Could apply incorrect GPT-5.2 rules to future models like gpt-5.20 if released.
   Recommendation: Use regex ^${key}(-|$) or exact match first, then prefix match only if no exact match exists.
   Status: 🟡 Needs review

2. Issue / Risk: DEBUG log in rosetta-ai.ts only covers Anthropic streams; GPT-5 OpenAI mappings lack similar debug logging despite complex parameter transformations.
   Impact: Harder to diagnose GPT-5-specific mapping failures in production.
   Recommendation: Extend debug logging to OpenAI GPT-5 paths (e.g., log support object and effective reasoning_effort, verbosity, sampling decision).
   Status: 🟡 Needs review

Language/Framework Checks

• TypeScript types correctly define ReasoningEffort ('none' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh') and Verbosity ('low' | 'medium' | 'high') in params.types.ts
• gpt5.support.ts imports ReasoningEffort from ../../types, but types/index.ts exports GenerateParamsProviderState—ensure types.ts re-exports ReasoningEffort (not visible in diff, but critical for build)
• getGpt5Support sorts keys by descending length (b.length - a.length) to prioritize longer prefixes (e.g., gpt-5.2 over gpt-5)—correct and necessary

Security & Privacy

• No sensitive data exposure or hardcoded secrets detected.
• Sampling parameter suppression for non-allowed models (temperature, top_p) prevents accidental leakage of user-controlled randomness into models where it’s disallowed (e.g., gpt-5, gpt-5.4).

Build/CI & Ops

• No changes to build scripts or dependencies—safe for deployment.
• New gpt5.support.ts file adds ~160 lines but is self-contained; no impact on bundle size beyond inclusion in dist.

Tests

• Unit tests added for core GPT-5 behaviors: default effort, sampling suppression, verbosity, prefix matching, fixed effort, and error cases.
• Missing test: verify verbosity is dropped for models without supportsVerbosity (e.g., gpt-5-codex)—add test case.
• Missing test: ensure maxTokens is preserved for unknown models but dropped for known ones (e.g., gpt-5.4-pro should drop maxTokens).

Approval Recommendation
Approve with caveats

• Address prefix-matching edge case in getGpt5Support (regex or stricter matching)
• Add missing verbosity/maxTokens test cases for unsupported models
• Confirm ReasoningEffort is exported from types/index.ts (verify in build)

[Copilot]

Pull request overview

Updates the OpenAI Chat Completions parameter mapping to be GPT‑5-family aware, introducing explicit reasoning/verbosity controls and model-specific capability rules.

Changes:

• Add reasoningEffort and verbosity to GenerateParams and plumb them through base param mapping.
• Implement a centralized GPT‑5 support table (getGpt5Support) and update OpenAIMapper to enforce/normalize GPT‑5 Chat Completions behavior (sampling, reasoning, verbosity, unsupported variants).
• Expand unit tests to cover GPT‑5 / GPT‑5.2 / dated variants / unsupported variants / unknown future variants.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/unit/core/mapping/openai.mapper.spec.ts	Adds test coverage for GPT‑5-family normalization and rejection behavior in Chat Completions mapping.
src/types/params.types.ts	Introduces ReasoningEffort / Verbosity types and exposes them on GenerateParams.
src/core/rosetta-ai.ts	Adds DEBUG-only logging of mapped Anthropic stream requests.
src/core/mapping/openai.mapper.ts	Implements GPT‑5 capability-aware mapping (reasoning, verbosity, sampling, unsupported model rejection).
src/core/mapping/gpt5.support.ts	New GPT‑5 capability table + exact/prefix lookup helper used by the OpenAI mapper.
src/core/mapping/common.utils.ts	Extends mapBaseParams to carry reasoningEffort and verbosity.
package.json	Bumps package version to 1.12.0.

---

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

You can also share your feedback on Copilot code review. Take the survey.