Kimi K2.6 ships model + agent: 4,000 tool calls per run

kimi just dropped k2.6 – and did it the new way: model + agent, shipped together

this is already the trend:

• gpt + openclaw. openai aligning openclaw to its models, turning it into a native execution layer
• minimax + hermes (maxhermes). minimax pairing its model with a managed hermes agent for distribution
• gemini + agent. google building the agent directly into the model experience

kimi just followed the same playbook – but went deep on execution.

model (k2.6):
• swe-bench pro: 58.6
• swe-bench multilingual: 76.7
• browsecomp: 83.2
• charxiv (python): 86.7
• math vision (python): 93.2\

agent (k2.6):
• 4,000+ tool calls per run
• 12+ hours continuous execution
• 300 parallel sub-agents
• 4,000 steps per run
• 100+ files from one prompt

this is no longer about shipping a better model. if you want to compete, you ship the system: model + agent = default.