Skip to content

v0.3.26 - Default to GPT-6 Luna and Claude Opus 5.5 (#905)

Choose a tag to compare

View all tags
@github-actions github-actions released this 22 Sep 19:48
· 8 commits to main since this release
v0.3.26
5b14df8
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

🌟 Summary

Release v0.3.26 updates the default AI models to GPT-6 Luna and Claude Opus 5.5, with matching pricing, reasoning, caching, documentation, and compatibility improvements.

📊 Key Changes

  • 🚀 New OpenAI defaults
    • General-purpose and PR review requests now default to gpt-6-luna, replacing gpt-5.6-luna.
    • Adds pricing support for gpt-6-luna and gpt-6-sol.
  • 🤖 Updated Anthropic PR review model
    • PR reviews now default to claude-opus-5-5, replacing claude-opus-5.
    • The general Anthropic model remains claude-sonnet-5.
  • 💰 Improved usage and cost tracking
    • GPT-6 models now support cache-write and long-context billing rules previously used for GPT-5.6 models.
    • Existing model pricing remains available so users can continue pinning older models.
  • 🧠 GPT-6 reasoning support
    • Reasoning-effort settings are now correctly passed to GPT-6 models.
    • Removes the unused temperature option from response APIs and callers, avoiding unsupported or unnecessary request parameters.
  • 📚 Documentation and release updates
    • Updates the English README, Chinese README, and GitHub Action metadata with the new defaults.
    • Removes an outdated Claude Sonnet pricing note.
    • Bumps the package version to 0.3.26.
  • ✅ Validation
    • All 173 tests passed.
    • Live requests using both new defaults returned successfully.

🎯 Purpose & Impact

  • ⚡ Better out-of-the-box performance: New installations automatically use the latest supported OpenAI and Anthropic review models.
  • 💸 More accurate cost controls: GPT-6 usage, caching, and long-context charges are now accounted for correctly, helping max_cost limits remain reliable.
  • 🛠️ Simpler API behavior: Removing the redundant temperature argument reduces configuration noise and prevents compatibility issues with reasoning-focused models.
  • 🔒 Backward compatibility: Users who explicitly select older models can continue using them because their pricing entries remain supported.
  • 📦 Action users should review defaults: Existing workflows that rely on automatic model selection will switch to GPT-6 Luna or Claude Opus 5.5 after upgrading. Users who need stable historical behavior can pin a model explicitly.

What's Changed

Full Changelog: v0.3.25...v0.3.26

Contributors

glenn-jocher
Assets 3
Loading