Aider 0.71 dropped within hours of Anthropic’s Claude 3.7 announcement. The release adds support for the new models (Claude 3.7 Opus, Sonnet, and Haiku) plus several quality-of-life improvements to architect mode and the diff edit format.

For aider users, this is a smooth upgrade. The Claude 3.7 models slot into existing config patterns; new features are opt-in.

The Claude 3.7 support

Three new models become available in aider:

• claude-3-7-opus-latest: highest capability, most expensive
• claude-3-7-sonnet-latest: middle tier (best balance for most users)
• claude-3-7-haiku-latest: fast and cheap

The naming follows Anthropic’s convention. Switch by updating .aider.conf.yml:

model: claude-3-7-sonnet-latest
weak-model: claude-3-7-haiku-latest

The improvements over 3.5:

• Sonnet 3.7 has better reasoning, comparable speed
• Opus 3.7 is the new flagship for hard tasks
• Haiku 3.7 is meaningfully faster while staying cheap

Architect mode refinements

Architect mode (introduced in 0.62, refined in 0.70) gets more polish:

Plan visibility. The architect’s plan now displays in a structured panel rather than inline chat text. Easier to read, easier to edit before execution.

Editor model auto-fallback. When the editor model produces a malformed diff, aider falls back to the architect model for that step. Reduces the “plan good, edit broken” failure mode.

Mid-plan checkpoints. For large architect plans, aider now pauses between major steps to confirm. You can adjust direction mid-execution rather than running the full plan.

These are small improvements. Combined, they make architect mode more pleasant to use for serious work.

Diff edit format improvements

The search-replace diff format that aider uses got tweaks:

• Better handling of files with mixed line endings
• More resilient parsing when the model emits slightly malformed blocks
• Improved error messages when diffs don’t apply

Less visible than the architect mode work, but reduces the friction of “the model did something close to right; aider failed to apply.”

What’s not new

A few things that didn’t make 0.71 but are commonly requested:

Native MCP support. Aider doesn’t speak MCP. Tools using MCP servers (Cline, Claude Code) have access to a growing ecosystem. Aider users have to use traditional integrations or workarounds.

Better cross-file refactoring. Aider’s strong on single-file edits and architect-mode multi-file plans. The middle ground (medium refactors with 3-5 files where architect mode is heavy) still has friction.

Native team features. Aider is single-user by design. Users wanting team features (shared config, audit logs, etc.) build their own workflows.

These are areas where competitors (Cline, Claude Code) have advantages. Aider remains focused on the single-user CLI experience.

Worth upgrading?

For aider users, yes. The upgrade is non-disruptive. Existing configs continue to work; new features are opt-in.

The Claude 3.7 model support alone justifies the upgrade if you use Claude. Even if you don’t, the diff format and architect mode improvements are worth the few minutes to update.

How aider’s evolution looks

Aider has been releasing steadily for over a year. The release cadence (2-4 weeks) is faster than most open-source CLI tools. Each release moves the product forward without breaking existing workflows.

For an open-source tool maintained by a small team, the pace is impressive. Aider has stayed competitive with VC-backed editor tools through consistent execution rather than feature leaps.

For users who like CLI workflows, prefer BYOK over subscriptions, or want git-native AI coding, aider remains the strongest option in its category. The 0.71 release reinforces that position without dramatic changes.

Update path

Standard pip install:

pip install --upgrade aider-chat

Or via pipx:

pipx upgrade aider-chat

For users on the latest Anthropic API key, the new models become available immediately. For users on older provider configs, no changes needed — your existing models continue to work.

What I’d watch in the next few releases

Based on the maintainers’ public discussion:

• Streaming output for the editor model in architect mode
• Better integration with various git workflow patterns (worktrees, submodules)
• Possibly some lightweight MCP-equivalent for tool integration

These are speculative. The aider team’s roadmap isn’t published in detail; releases tend to surprise pleasantly.

The bigger picture

Aider’s continued strong releases matter for the AI coding ecosystem. As venture-backed tools (Cursor, Cline, Windsurf) become dominant, having a strong open-source alternative provides:

• Pricing pressure (BYOK is cheaper than subscriptions for some users)
• Tooling alternatives for users with privacy or compliance constraints
• A reference implementation for ideas the commercial tools may adopt

The 0.71 release isn’t a category-defining event. It’s a steady step that keeps aider as a viable choice. That continuity is the value.