Ollama v0.34.4 applies structured outputs on thinking models in a single pass
GitHub·low signal
The release (2026-09-23) says JSON-schema outputs on reasoning models are now faster and more reliable. It also speeds up Qwen 3.8 prompt processing on Apple Silicon, picks per-image resolution for Gemma 4, and fixes intermittent 'model not found' errors with large local libraries. If you run structured extraction on local thinking models, upgrade.