Adds support for the full GPT-5 model line, Grok-4 (via xai/grok-4 and openrouter/x-ai/grok-4), gemini/gemini-2.5-flash-lite-preview-06-17, and openrouter/moonshotai/kimi-k2. Rest is small UX polish and fixes, plus a litellm bump to 1.75.0.
Adds Gemini 2.5 Pro/Flash with thinking-token support and Responses-API models such as o1-pro and o3-pro, and repoints the flash and gemini aliases to the 2.5 models — so existing configs using those aliases will silently target a different model. Also adds several CLI flags (--add-gitignore-files, --commit-language, PostHog analytics options), MATLAB and Clojure repo-map support, and turns Co-authored-by commit attribution on by default.
Aider added Claude Sonnet 4 and Opus 4 support across providers and repointed the sonnet and opus aliases at those newer versions, so existing configs using the short aliases will now hit different models. Also adds a Vertex Gemini 2.5 Flash preview model, automatic GitHub Copilot token refresh, and a fix for .gitignored files added on the command line.
Aider dropped support for Python 3.9 — an environment-level break for anyone still on it. It also repointed the gemini and gemini-2.5-pro aliases at the 05-06 preview models, added qwen3-235b support, and enabled thinking_tokens / reasoning_effort for OpenRouter models.
Aider added support for the GPT-4.1 family (including mini and nano) and the xAI Grok 3 models via both xAI and OpenRouter. It introduces a patch edit format for GPT-4.1 plus three editor-* edit formats, and fixes automatic edit-format selection in architect mode.
Aider will now OAuth against OpenRouter when started with no model and no keys, and picks a default model based on which provider credentials it finds — OpenRouter free/paid tier, GEMINI_API_KEY, or VERTEXAI_PROJECT. Switching models with /model now also switches the edit format if you were on the old model's default.
Aider adds support for Gemini 2.5 Pro and DeepSeek V3 0324, plus a new /context command that works out which files a request needs edited. Also adds an "overeager" mode for Claude 3.7 Sonnet aimed at keeping the model inside the requested scope.
Aider now validates whether a model actually supports --reasoning-effort and --thinking-tokens and ignores them if not, gated by a new --check-model-accepts-settings flag. Adds /editor-model and /weak-model commands to switch model roles mid-session, thinking-token support for OpenRouter Sonnet 3.7, and a --git-commit-verify flag controlling whether git commit hooks are bypassed.
Aider switches to tree-sitter-language-pack, adding linter support for 130 more languages and repo-map support for 20 more. Adds /think-tokens and /reasoning-effort commands for runtime control of reasoning budget, and turns on --auto-accept-architect by default so architect-format changes apply without confirmation.
Aider adds a --thinking-tokens budget control and displays reasoning content returned by models. Claude 3.7 Sonnet is now supported on OpenRouter, Bedrock and Vertex AI and becomes the default model on OpenRouter; the old model shortcut flags (--4o, --opus) are deprecated in favour of --model.
Basic Claude 3.7 Sonnet support lands; --model sonnet now selects 3.7, with thinking support stated as coming later. Rest of the release is syntax support and bug fixes.