Ollama: MLX Compatibility Push and Responses Fixes
Most activity tightens MLX import and runtime support for Qwen and Bonsai checkpoints that previously failed or silently misbehaved. A merged Responses change improves handling of provider switches, plus new launch wiring for Mux and Claude Code.
Duration: PT2M40S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-10-19T13:00:39Z
- Audio duration: PT2M40S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Hello, it's October 19th, 2026, and this is your Ollama developer briefing.
The headline is tolerance for messy real-world checkpoints. Most of today's work makes MLX models import successfully and run correctly where they previously failed, loaded wrong data, or fell back to slow paths.
First, MLX import and runtime hardening. For example, PR 19135 handles older quantization configs that record only bits and group size without naming the mode, like the Qwen 3 8-billion 4-bit pack cited in the report. Without it, weights imported with no quant type and the runner had to guess from tensor shapes. And…
Second, OpenAI compatibility. Merged PR 19003, commit 6787222, lets Responses history survive a provider switch by omitting another provider's opaque compaction item, preserving supplied input in order, and expanding the newest native Ollama checkpoint, rejecting only when nothing usable remains. PR 19114 fixes a…
Finally, agent launch. PR 19125 adds ollama launch support for Mux, detecting the tool on path and writing a dedicated endpoint while preserving other endpoints and existing defaults. PR 19119 passes the selected cloud model's context length into Claude Code to avoid the…
What's…
Nearby episodes from Ollama
- Remote Cancel Fixes and Wider Model Support
- Weekly Recap - Decision Models Go Live, MLX Gains, and API Polish
- Rendering Transparency and Response Fidelity Fixes
- Proxy Transfers and API Resilience
- Model Compatibility Catch-Up and Windows Reliability
- Responses Compatibility and Clearer Errors
- Decision Models Go Usable, MLX Speeds Up on CUDA
- Decision Scoring Matures as Inference Hardens