Aider v0.77.0: 130 New Languages & Thinking Tokens
Aider v0.77.0 adds 130 new languages, thinking token controls for Claude, and auto-accept architect mode. Aider wrote 72% of this release.
TL;DR
- 130 new languages with linter support via tree-sitter-language-pack
- New `/think-tokens` and `/reasoning-effort` commands for Claude's extended thinking
- Auto-accept architect coder format by default; support for Cohere and Gemini models
New
- Language pack expansion — 130 new languages with linter support and 20 with repo-map support by adopting tree-sitter-language-pack.
- /think-tokens command — Set thinking token budget with human-readable formats (8k, 10.5k, 0.5M); displays current setting when called without arguments.
- /reasoning-effort command — Control model reasoning level; shows current setting when called without arguments.
- --auto-accept-architect flag — Automatically accept changes from architect coder format without confirmation (enabled by default).
- New model support — Added cohere_chat/command-a-03-2025 and gemini/gemma-3-27b-it.
- Thinking tokens in model info — Display of thinking token budget and reasoning effort in model information output.
- --thinking-tokens argument — Now accepts string values with human-readable formats instead of raw numbers.
Fixed
- Bare `/drop` command now preserves original read-only files provided via args.read.
- Default model no longer overridden by deprecated `--shortcut` switches when model already specified in command line.
- AutoCompleter now requires 3 characters for autocompletion to reduce noise.
- JSONDecodeError handling when loading model cache file.
- GitCommandError handling when retrieving git user configuration.
- File watcher now handles restricted files gracefully with ignore_permission_denied option.
Update: pip install --upgrade aider-ai
Source: Aider