Aider 0.72 dropped this week with smoother integration of OpenAI’s o3-mini for the architect role in architect mode. The release also fixes some long-standing issues with multi-file diff application.

What’s new

o3-mini for architect:

# .aider.conf.yml
model: claude-3-5-sonnet
weak-model: claude-3-5-haiku
architect-model: o3-mini
editor-model: claude-3-5-haiku

This was already supported in 0.71 but with rough edges. 0.72 polishes the integration:

• Better handling of o3-mini’s reasoning output
• Cleaner integration with editor models from other providers
• Improved error messages when the model is rate-limited

Multi-file diff fixes:

Earlier versions of aider sometimes failed to apply multi-file diffs cleanly. The model would produce a sequence of search-replace blocks; aider would parse them but miss a block silently. The fix in 0.72 makes parsing more robust and surfaces failures clearly.

Reduced repo map size for large repos:

The repo map for a million-line codebase has been a token hog. 0.72 introduces smarter scoping — the repo map now prioritizes files near recent edits rather than including everything.

The o3-mini integration

For users curious about the architect/editor split with cross-provider models:

The architect (o3-mini in this config) plans the change. The editor (Claude Haiku in this config) translates the plan into specific diffs.

The trick is that the architect produces text, not diffs. The editor reads the architect’s text plus the actual files, then produces diffs.

This mostly works but had failure modes:

• O3-mini’s plans sometimes referenced specific code patterns that the editor couldn’t translate
• Token usage was hard to predict because reasoning tokens are billed differently
• The plan/edit handoff lost some context

0.72’s improvements address these. The integration is smoother in practice.

When this matters

For most users, default Claude across architect and editor is fine. The cross-provider mix is for specific cases:

• Users with strong preferences for one provider’s reasoning vs another’s diff format
• Users optimizing for cost (different price points across providers)
• Users with rate limits at one provider but not the other

For these cases, the smoother integration is a real improvement.

Other changes

Smaller improvements in 0.72:

• Faster startup on large repos (better caching)
• Better progress indicators during long operations
• Improved handling of network errors (cleaner retries)
• Updated model defaults for Claude 3.7 family

None individually are dramatic. Cumulatively, the experience is incrementally smoother.

Worth upgrading?

Yes. The upgrade is non-disruptive. Existing configurations continue working. New features are opt-in.

For users on architect mode, the o3-mini integration is the main value. For users not on architect mode, the multi-file diff fixes are the value.

Aider’s continued release pace

Aider has shipped 12 releases in the past year. The cadence is steady; quality is consistent. For an open-source project maintained by a small team (mostly the original author plus a handful of contributors), the pace is impressive.

The product evolves without breaking changes. Existing users keep working; new users get newer features. This is rare in fast-moving software.

The market position

Aider remains the strongest CLI-based AI coding tool that’s also fully open source. Claude Code is closed-source; Codex CLI is closed-source; Continue’s CLI is less mature.

For users who:

• Prefer CLI over editor-based workflows
• Want BYOK with provider flexibility
• Like the git-native commit pattern
• Value open source

…aider remains the natural choice. The 0.72 release reinforces this position without dramatic changes.

Update path

Standard pip install:

pip install --upgrade aider-chat

Or pipx:

pipx upgrade aider-chat

For most users, the upgrade should be transparent. Existing configs continue working.

What’s next for aider

Based on the maintainer’s public discussion:

• Better integration with various git workflows
• More refinements to the architect mode
• Possibly some lightweight team features
• Continued model support as new releases arrive

The roadmap isn’t formal. Aider has been able to release frequently because there’s no marketing pressure or quarterly milestones. The releases happen when they’re ready.

For users, this means steady improvement without dramatic shifts. For users who like predictability, aider’s pace is comfortable.

The broader observation

The CLI agent category is now competitive with editor-based AI tools for many workflows. Claude Code, Codex CLI, aider, Continue CLI — each has strengths. The category is mature enough that real evaluation is warranted rather than defaulting to editor-based tools.

Aider’s specific position: open source, BYOK-friendly, git-native, mature. Worth knowing about even if you ultimately use other tools. The patterns aider established (architect mode, repo map, edit format) influence the broader ecosystem.

For 2026, expect continued aider development at the steady pace. The 0.72 release is one of many that compound into substantive improvement over time.