Aider

• Big upgrade in programming languages supported by adopting tree-sitter-language-pack.
  • 130 new languages with linter support.
  • 20 new languages with repo-map support.
• Added /think-tokens command to set thinking token budget with support for human-readable formats (8k, 10.5k, 0.5M).
• Added /reasoning-effort command to control model reasoning level.
• The /think-tokens and /reasoning-effort commands display current settings when called without arguments.
• Display of thinking token budget and reasoning effort in model information.
• Changed --thinking-tokens argument to accept string values with human-readable formats.
• Added --auto-accept-architect flag (default: true) to automatically accept changes from architect coder format without confirmation.
• Added support for cohere_chat/command-a-03-2025 and gemini/gemma-3-27b-it
• The bare /drop command now preserves original read-only files provided via args.read.
• Fixed a bug where default model would be set by deprecated --shortcut switches even when already specified in the command line.

• Improved support for thinking/reasoning models:

  • Added --thinking-tokens CLI option to control token budget for models that support thinking.
  • Display thinking/reasoning content from LLMs which return it.
  • Enhanced handling of reasoning tags to better clean up model responses.
  • Added deprecation warning for remove_reasoning setting, now replaced by reasoning_tag.

• Aider will notify you when it's completed the last request and needs your input:

  • Added notifications when LLM responses are ready with --notifications flag.
  • Specify desktop notification command with --notifications-command.

• Added/improved support for many models/providers:

  • Added support for QWQ 32B.
  • Support for DeepSeek V3 free on OpenRouter: --model openrouter/deepseek/deepseek-chat:free.
  • Added support for Claude 3.7 Sonnet models on OpenRouter, Bedrock and Vertex AI.
  • Updated default model to Claude 3.7 Sonnet on OpenRouter.
  • Added support for GPT-4.5-preview model.
  • Added support for Claude 3.7 Sonnet:beta on OpenRouter.

• Basic support for Claude 3.7 Sonnet
  • Use --model sonnet to use the new 3.7
  • Thinking support coming soon.
• Bugfix to /editor command.
• Downgrade streamlit dependency to avoid threading bug.
• Added support for tree-sitter language pack.
• Added openrouter/o3-mini-high model configuration.
• Added build.gradle.kts to special files for Kotlin project support, by Lucas Shadler.
• Prevent more than one cache warming thread from becoming active.
• Fixed continuation prompt ". " for multiline input.
• Added HCL (Terraform) syntax support, by Warren Krewenki.
• Have o1 & o3-mini generate markdown by sending the magic "Formatting re-enabled." string.
• Bugfix for multi-line inputs, which should not include the ". " continuation prompt.
• Aider wrote 46% of the code in this release.

• Dynamically changes the Ollama context window to hold the current chat.
• Better support for o3-mini, DeepSeek V3 & R1, o1-mini, o1 via secondary API providers.
• Remove <think> tags from R1 responses for commit messages (and other weak model uses).
• Can now specify use_temperature: <float> in model settings, not just true/false.
• The full docker container now includes boto3 for Bedrock.
• Docker containers now set HOME=/app which is the normal project mount-point, to persist ~/.aider.
• Bugfix to prevent creating incorrect filenames like python, php, etc.
• Bugfix for --timeout
• Bugfix so that /model now correctly reports that the weak model is not changed.
• Bugfix so that multi-line mode persists through ^C at confirmation prompts.
• Watch files now fully ignores top-level directories named in ignore files, to reduce the chance of hitting OS watch limits. Helpful to ignore giant subtrees like node_modules.
• Fast startup with more providers and when model metadata provided in local files.
• Improved .gitignore handling:
  • Honor ignores already in effect regardless of how they've been configured.
  • Check for .env only when the file exists.

• Full support for o3-mini: aider --model o3-mini
• New --reasoning-effort argument: low, medium, high.
• Improved handling of context window size limits, with better messaging and Ollama-specific guidance.
• Added support for removing model-specific reasoning tags from responses with remove_reasoning: tagname model setting.
• Auto-create parent directories when creating new files, by xqyz.
• Support for R1 free on OpenRouter: --model openrouter/deepseek/deepseek-r1:free
• Enforce user/assistant turn order to avoid R1 errors, by miradnanali.
• Case-insensitive model name matching while preserving original case.
• Harden against user/assistant turn order problems which cause R1 errors.
• Fix model metadata for openrouter/deepseek/deepseek-r1
• Aider wrote 69% of the code in this release.

https://aider.chat/HISTORY.html

• Support for DeepSeek R1.
  • Use shortcut: --model r1
  • Also via OpenRouter: --model openrouter/deepseek/deepseek-r1
• Added Kotlin syntax support to repo map, by Paul Walker.
• Added --line-endings for file writing, by Titusz Pan.
• Added examples_as_sys_msg=True for GPT-4o models, improves benchmark scores.
• Bumped all dependencies, to pick up litellm support for o1 system messages.
• Bugfix for turn taking when reflecting lint/test errors.
• Fix permissions issue in Docker images.
• Added read-only file announcements.
• Bugfix: ASCII fallback for unicode errors.
• Bugfix: integer indices for list slicing in repomap calculations.
• Aider wrote 52% of the code in this release.

• Prompts to help DeepSeek work better when alternating between /ask and /code.
• Streaming pretty LLM responses is smoother and faster for long replies.
• Streaming automatically turns of for model that don't support it
  • Can now switch to/from /model o1 and a streaming model
• Pretty output remains enabled even when editing files with triple-backtick fences
• Bare /ask, /code and /architect commands now switch the chat mode.
• Increased default size of the repomap.
• Increased max chat history tokens limit from 4k to 8k.
• Turn off fancy input and watch files if terminal is dumb.
• Added support for custom voice format and input device settings.
• Disabled Streamlit email prompt, by apaz-cli.
• Fixed lint command handling of nested spaced strings, by Aaron Weisberg.
• Added token count feedback when adding command output to chat.
• Improved error handling for large audio files with automatic format conversion.
• Improved handling of git repo index errors, by Krazer.
• Improved unicode handling in console output with ASCII fallback.
• Added AssertionError to git error handling.
• Fixed file export path in voice format conversion.

• Full support for o1 models.
• Watch files now honors --subtree-only, and only watches that subtree.
• Improved prompting for watch files, to work more reliably with more models.
• New install methods via uv, including one-liners.
• Support for openrouter/deepseek/deepseek-chat model.
• Better error handling when interactive commands are attempted via /load or --load.
• Display read-only files with abs path if its shorter than rel path.
• Ask 10% of users to opt-in to analytics.
• Bugfix for auto-suggest.
• Gracefully handle unicode errors in git path names.
• Fix for gemini model names in model metadata.
• Show hints about AI! and AI? when user makes AI comments.
• Support for running without git installed.
• Improved environment variable setup messages on Windows.
• Aider wrote 74% of the code in this release.

https://aider.chat/HISTORY.html

• Watch files improvements:

  • Use # ... AI? comments to trigger aider and ask questions about your code.

  • Now watches all files, not just certain source files.

  • Use # AI comments, // AI comments, or -- AI comments to give aider instructions in any text file.

• Full support for Gemini Flash 2.0 Exp:

  • aider --model flash or aider --model gemini/gemini-2.0-flash-exp

• New --multiline flag and /multiline-mode command makes ENTER a soft newline and META-ENTER send the message, by @miradnanali.

• /copy-context <instructions> now takes optional "instructions" when copying code context to the clipboard.

• Improved clipboard error handling with helpful requirements install info.

• Ask 5% of users if they want to opt-in to analytics.

• /voice now lets you edit the transcribed text before sending.

• Disabled auto-complete in Y/N prompts.

• Aider wrote 68% of the code in this release.