RC RANDOM CHAOS

Moonshot's Kimi K3.1 Pro Rivals Top Models, Opens Weights Soon

· via Simon Willison

Original source

Introducing Kimi K3.1 Pro

Simon Willison →

Moonshot released Kimi K3.1 Pro on its API with full model weights arriving by early November. Pricing stands at $2 per million input tokens and $10 per million output tokens—competitive with GPT-6.1 Sol and half the cost of Claude Opus 5.5. The service includes a faster variant delivering 2-3x speedup at double pricing, though token cache hits maintain the standard rate.

In benchmarks, Kimi performs alongside Claude Opus 5.5 and GPT-6 Astra, with wins on some tests and near-parity on others. Artificial Analysis rated it 55, surpassing Claude Fable 5.1 and GPT-6 Astra (both 53), with only Claude Opus 5.5 (58) ranking higher. The significance: this is the first model scheduled for open-source release to achieve this caliber of frontier performance.

The model supports three reasoning modes—low, medium, and max. In testing, these levels generate distinct outputs, providing users with configurable options upon public release.

Read the full article

Continue reading at Simon Willison →

This is an AI-generated summary. Read the original for the full story.