feat: update model registry to April 2026 (TMC-92)#431
Conversation
Migrate from pip/requirements.txt workflow to uv-native dependency management with devenv for reproducible development environments. Bumps Python target to 3.14 and requires-python to >=3.12. - Add .python-version as single source of truth for Python version - Add [dependency-groups] dev/lint to pyproject.toml, replacing requirements-dev.txt as canonical dev deps - Add uv.lock for deterministic dependency resolution - Add secretspec.toml + devenv.local.yaml for declarative 1Password secret management via SecretSpec - Modernize code_quality_checks.sh and run_integration_tests.sh to use uv sync and direct tool invocation instead of .pal_venv/pip - Update ruff target-version to py312, black to py312/py313 - Update .pre-commit-config.yaml excludes for .devenv/ - Update isort skip_glob for .devenv/ - Apply ruff 0.15 auto-fixes (pyupgrade, asyncio.TimeoutError -> TimeoutError, etc.) Co-Authored-By: Claude <noreply@anthropic.com>
Remove devenv.local.yaml (contains personal 1Password vault ID) from tracking and add devenv artifacts to .gitignore. Un-ignore .python-version so it's tracked as the Python version source of truth. Co-Authored-By: Claude <noreply@anthropic.com>
Commit standalone devenv.nix, devenv.yaml, and .envrc so anyone can clone the repo and get a working environment with `devenv shell`. No personal config - devenv.local.yaml (for secret providers like 1Password) stays gitignored. SecretSpec secrets marked optional so devenv starts without API keys (needed for linting/tests, not for API calls). Co-Authored-By: Claude <noreply@anthropic.com>
Add GPT-5.4 family (5.4, 5.4 Pro, 5.4 Mini, 5.4 Nano) and GPT-5.3 Codex. Add Gemini 3.1 Pro Preview, 3 Flash Preview, and 3.1 Flash Lite Preview. Remove dead/retired models: gpt-4.1, gemini-3-pro-preview (already 404), gemini-2.0-flash, gemini-2.0-flash-lite. Mark Gemini 2.5 models as deprecated (June 17 shutdown). Fix Gemini provider to branch on thinking_level (3.x enum API) vs thinking_budget (2.x numeric API) - without this, 3.x models 400-error on thinking requests. Fix preference selection to exclude flash-lite from flash candidates. Fix registry to use canonical names only in preference selection (was including aliases, causing sort-order bugs). Co-Authored-By: Claude <noreply@anthropic.com>
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces significant updates to the model provider configurations, including the adoption of Gemini 3.1 and GPT-5.4 models, updates to the development environment using devenv and uv, and a transition to modern Python type hinting syntax (e.g., | None). The review identified potential fragility in the Gemini 3.x model identification and Pro model detection logic, as well as a need for better error messaging in the code quality check script.
| if level == "MINIMAL" and "pro" in resolved_model_name: | ||
| level = "LOW" |
There was a problem hiding this comment.
The check if level == "MINIMAL" and "pro" in resolved_model_name: is fragile. If a model name contains 'pro' but is not a Pro model, this logic will fail. Consider using a more robust way to identify Pro models, such as checking against a list of known Pro model names or a capability flag.
| if ! command -v uv &> /dev/null; then | ||
| echo "Error: uv not found. Activate devenv first (cd into repo with direnv, or run devenv shell)." | ||
| exit 1 | ||
| fi |
There was a problem hiding this comment.
The script uses command -v uv to check for uv. If uv is not found, it exits with an error. It is better to provide instructions on how to install uv or how to set up the environment properly, rather than just stating it is not found.
| def _is_gemini_3x(model_name: str) -> bool: | ||
| """Return True if the model uses the Gemini 3.x thinking_level API.""" | ||
| return model_name.startswith("gemini-3") |
There was a problem hiding this comment.
The _is_gemini_3x method uses model_name.startswith("gemini-3"). This might incorrectly identify future models (e.g., 'gemini-30-pro') as 3.x models. It is safer to use a more specific check, such as model_name.startswith(("gemini-3.", "gemini-3-")).
1 participant
Summary
thinking_levelenum for 3.x models instead ofthinking_budget(breaking API change - 3.x models 400-error without this)Test plan
./code_quality_checks.shpasses (ruff, black, isort)google-genaiSDK supportsThinkingConfig(thinking_level='HIGH')Closes TMC-92
🤖 Generated with Claude Code