Moonshot AI’s Kimi K3.1 Pro has scored 55 on the Artificial Analysis Intelligence Index, an 11-point gain over its predecessor and enough for fourth place. It sits three points behind Claude Opus 5.5, with Claude Sonnet 5.5 and Muse Wonder Full between them.
The new model keeps K3’s architecture at 2.8 trillion total parameters with 104 billion active at a time, and supports a one-million-token context window plus text and images. Moonshot attributes the gains to additional post-training. A Highspeed edition is advertised as two to three times faster.
K3.1 Pro is available through Kimi’s consumer products, coding tool and API. Input now costs $2 per million tokens, output $10 and cached input $0.20, all one-third below the previous rates. Moonshot says it plans to release model weights by November 4; until that happens, the open-weight claim remains a plan rather than an available download.
Benchmark figures also need context. Many results come from Moonshot’s own testing, including a fall in claim-level hallucinations from 23.4% to 8.3% on its internal stress test. The company removed some offensive cybersecurity training data from the planned open release, reducing performance on its vulnerability benchmark. Independent use will be needed to test how the reported gains translate to everyday workloads.