Kimi K2.5

Kimi K2.5 — coming soon

Moonshot AI's long-context reasoner. Two-million token context means you can stuff entire codebases, book-length documents, or multi-hour meeting transcripts into a single call — no RAG gymnastics.

Track record (public benchmarks)
- Top-tier math reasoning (competitive with o3-mini)
- Exceptional long-document comprehension (needle-in-a-haystack near-perfect at 2M)
- Native Agent Swarm v1 — the model itself can spawn up to 100 sub-agents
- Strong coding performance, slightly behind Qwen3 235B on short tasks

Availability: on the Gonka roadmap — lights up once validators onboard the weights. The platform's agent-swarm block already targets it: the day K2.5 goes live, just pass model=kimi-k2.5.