Side by side
The figures come from Moonshot's announcement and pricing page as reported on October 23, 2026.
- Context window: 1M tokens on K3.1 Pro.
- API price per 1M tokens: K3.1 Pro $2 input / $10 output; K3 $3 / $15.
- Kimi Code Bench 3.0 Long-horizon (internal): K3.1 Pro 61.2%, K3 55.95%.
- Financial modeling and research (internal): K3.1 Pro is 7.2 and 11.6 points higher.
- Hallucination stress test (internal): K3.1 Pro 8.3%, K3 23.4%.
What stayed the same
K3.1 Pro keeps K3's mixture-of-experts design and multimodal input. Moonshot's gains come from post-training: larger-scale reinforcement learning, better long-context inference and stricter environment checks.
Which one to use
Use K3.1 Pro for new work: it is cheaper and scores higher on every benchmark Moonshot published. Choose Highspeed when response time matters more than cost, such as interactive coding.