October 26, 2026

--tool-strict-level in vLLM lifts Kimi tool-call schema accuracy from 92% to 100%

On vLLM 0.28, Kimi K2.6 broke the schema in 2 out of 25 tool calls. A single flag at server launch brought accuracy to 100%.

Source

One email a week - only what changes how you work. Subscribe

October 26, 2026

On vLLM 0.28, Kimi K2.6 broke the schema in 2 out of 25 tool calls. A single flag at server launch brought accuracy to 100%.

Source

One email a week - only what changes how you work. Subscribe