Aider v0.78.0: Thinking Tokens & Model Switching
Aider v0.78.0 adds thinking token support for reasoning models, new commands to switch model types mid-session, and smarter model validation that ignores unsupported settings automatically.
TL;DR
- Thinking tokens support for OpenRouter Sonnet 3.7 — reasoning models now work properly
- New `/editor-model` and `/weak-model` commands to switch between model types mid-session
- Model validation now ignores unsupported settings automatically (with override flag available)
New
- Thinking tokens for OpenRouter Sonnet 3.7 — Added support for reasoning tokens on OpenRouter's Sonnet 3.7 model.
- /editor-model and /weak-model commands — Switch between Editor Model and Weak Model types without restarting, contributed by csala.
- Model setting validation — Automatically ignores `--reasoning-effort` and `--thinking-tokens` if your model doesn't support them.
- --check-model-accepts-settings flag — Override validation to force unsupported settings (default: true).
- --git-commit-verify flag — Control whether git commit hooks are bypassed (default: False).
- AWS_PROFILE support for Bedrock — Use AWS profiles instead of explicit credentials, contributed by lentil32.
- Enhanced --aiderignore paths — Now resolves both absolute and relative paths, contributed by mopemope.
- openrouter/google/gemma-3-27b-it model — New model option available.
- Vi-like enter behavior in multiline mode — Better navigation when in vi normal mode, contributed by Marco Mayer.
- Improved code block rendering — Better padding in markdown output using NoInsetMarkdown.
Fixed
- Autocompletion for `/ask`, `/code`, and `/architect` commands — Fixed missing completions, contributed by shladnik.
- Ollama compatibility — Bumped litellm dependency to pickup Ollama fix.
- Platform information handling — Gracefully handles retrieval errors.
Update: pip install --upgrade aider-ai
Source: Aider
