@magnitudedev/cli@0.2.5 @magnitudedev/cli@0.2.6
ChatGPTNextWeb/NextChatOct 3, 2026 → Oct 5, 20262 days between releases
What's New
0.2.6 is a bug-fix patch addressing local-model tool integration, Mac GPU stability, and long-request timeouts, with no new features.
Improvements
- Mac GPU stability and performance improved across M1, M2, and M5 hardware.
- Further performance tuning for Qwen models on top of the 0.2.5 MoE and prompt-processing gains.
- Long-running local model requests now tolerate extended generation and prompt processing without connection timeouts.
Bug Fixes
- Codex now correctly sends tool results back to local models, accepting replayed reasoning items with null content or summary as empty.
- OpenCode sessions no longer fail with an 'assistant content is required' error after a step that produced no output, across both Chat Completions and Anthropic APIs.
- A headless `magnitude serve` that fails to start now reports the underlying error instead of crashing with EBADF.
- Long requests to local models no longer fail with a 502 after roughly five minutes.
- Windows engine crashes caused by invalid JSON in chat templates are fixed.
Upgrade Notes
This is a drop-in patch upgrade from 0.2.5 with no configuration or migration steps required. Restart any long-running `magnitude serve` processes to pick up the timeout and startup-error fixes. Teams running local models on M1/M2/M5 Macs or Windows should see the most benefit.
The Two Releases
@magnitudedev/cli@0.2.5
Oct 3, 2026· 19 days ago
Significant performance gains for mixture-of-experts models and prompt processing on newer Macs, alongside fixes for loading models on M1 and M2.
Full release notes@magnitudedev/cli@0.2.6
Oct 5, 2026· 17 days ago
This release focuses on fixing critical bugs related to local model tool integration, Mac hardware compatibility, and performance optimizations for Qwen models.
Full release notes