Once Qwen3-VL is stable on Ollama, swap `qwen2.5vl:7b` / `qwen2.5vl:32b` for the Qwen3-VL equivalents in `src/ollama_runtime.py` (and the GPU-tier table in `docs/models.md`).
Validate by re-running the golden integration tests (issue #16); accept the swap only if accuracy is at least as good.
Once Qwen3-VL is stable on Ollama, swap `qwen2.5vl:7b` / `qwen2.5vl:32b` for the Qwen3-VL equivalents in `src/ollama_runtime.py` (and the GPU-tier table in `docs/models.md`).
Validate by re-running the golden integration tests (issue #16); accept the swap only if accuracy is at least as good.