Tools
vLLM's chat CLI finally streams reasoning deltas instead of going silent while a model thinks
PR #57045 by mgoin, merged 2026-09-16, makes `vllm chat` read the `reasoning` and `reasoning_content` fields so reasoning-parser output streams as it arrives. Before the change the CLI showed nothing during the entire thinking phase, which looked like a hang on long reasoning traces. Small change, but it is the difference between a usable and an unusable interactive smoke test for reasoning models served locally.
Source
↳ Follow the thread