I had the pleasure of chance encounters through LMArena yesterday and although I am not fully convinced yet that it actually is equal to gemini-3 or claude-4.5-opus across the board, it's performing pretty well!

optimism

Is it safe?
I don't trust Chineese brands.

orto

The main innovation is Agent Swarm mode. In complex tasks, the model manages a swarm of 100 subagents and makes up to 1,500 tool calls.

In coding, according to benchmarks, Kimi K2.5 performs at the level of Gemini 3 Pro.

https://www.kimi.com/

Trial access to Agent Swarm is only available on higher-tier plans.

The code and weights are available on HuggingFace.

The main innovation is Agent Swarm mode. In complex tasks, the model manages a swarm of 100 subagents and makes up to 1,500 tool calls.

In coding, according to benchmarks, Kimi K2.5 performs at the level of Gemini 3 Pro.

You can try it for free here: https://www.kimi.com/
Trial access to Agent Swarm is only available on higher-tier plans.

The code and weights are available on HuggingFace.