Ollama: Qwen3.5 Gets Untangled, and Small Bugs With Big Blast Radius
Two PRs fix Qwen3.5 model family misidentification that leaked broken output into chat, while three smaller fixes address a boolean config bug that could silently flip feature flags, a backwards error message, and an incomplete API schema.
Duration: PT2M44S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-07-09T13:01:06Z
- Audio duration: PT2M44S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good morning. It's July 9th, 2026, and today's briefing is about precision — in model matching, in config parsing, and in error messages that need to say what they mean.
The headline story is Qwen3.5. Because Qwen3.5 and Qwen3-Next architecture strings both contain the substring "qwen3," the system was matching them to the generic Qwen3 parser and the Qwen3-coder renderer. The problem: that renderer's template doesn't frame the thinking block, so an empty think tag leaked straight…
Second theme: small bugs with outsized consequences. PR 17087 fixes a boolean config parser that returned true on a parse error instead of falling back to the documented default — meaning a typo in an environment variable could silently turn a feature on. PR 17093 fixes an error message during model import that had…
Rounding out the day: PR 17088 adds PCI-based detection for Intel's Battlemage Arc GPUs on Linux, fixing a case where Intel-only hosts were treated as CPU-only. PR 17095 patches two dependency libraries to clear three reachable security advisories. And PR 17092 fixes the API documentation so it correctly reflects…
What to remember: if you're touching model family detection, check for shared…
That…
Nearby episodes from Ollama
- Model Behavior Bugs and Blob Security Hardening
- Launch Gets Smarter, Model Support Gets Wider
- Cleaning Up the Scheduler's Edge Cases
- Weekly Recap - Rebuilding the Model Pipeline and Tightening the Guardrails
- Truth in Reporting
- MLX Create Pipeline Rewrite Lands
- Agent Harness Lands, Hardware Support Gets a Cleanup
- Gemma 4 Support and Platform Improvements