What changed
Moonshot AI released Kimi K2 Thinking, a one-trillion-parameter mixture-of-experts model with native INT4 weights, a 256K context window, and training for interleaved reasoning and tool calls.
Why it matters
Open-weight models were being trained for persistent tool use rather than single-turn reasoning, narrowing the architecture gap with proprietary research and coding agents.
What to do
Test it on multi-step research and coding traces with deterministic tool schemas; inspect tool-call failure, context drift, token use, and recovery after a bad observation.
Caveat
The headline claim of hundreds of sequential tool calls and the benchmark results were Moonshot-reported. The model remains demanding to serve despite INT4 weights.