Bite-sized AI for curious minds...
Moonshot AI’s latest frontier model for coding and agents
Moonshot AI unveiled Kimi K3 in July 2026, and reporting says it is a 2.8 trillion-parameter open-source model. Coverage says it performs especially well on coding, web development, and agent benchmarks, positioning it as a major new Chinese model release.
Moonshot AI released Kimi K3.1 Pro on its API today, but the model is not yet open weights. The company promises full model weights will be released by November 4, 2026. A 'Highspeed' variant runs 2-3x faster at twice the price, though input token cache hits are priced the same. This gives developers a faster option for latency-sensitive tasks while keeping costs predictable for cached inputs.
Moonshot AI's Kimi K3.1 Pro scored 55 points on the Artificial Analysis Intelligence Index, an 11-point jump over its predecessor that places it fourth overall. Only three Anthropic and Meta models rank higher, and if Moonshot releases the weights as planned, the gap between open and proprietary models shrinks to just three points. The 11-point gain in a single generation signals that Chinese labs are iterating faster than Western counterparts on benchmark performance. This puts Kimi K3.1 Pro within striking distance of Claude Opus 5.5, which leads the index. The planned open-weight release would make it the highest-scoring open model available, directly challenging Meta's Llama and Anthropic's closed API.
Moonshot AI released Kimi K3.1 Pro on October 24, 2026, the first model in its new K3.1 series. The company calls it its most capable 3T-class open-weight model to date, featuring a 1-million-token context window. A faster companion variant, Kimi K3.1 Pro Highspeed, launched alongside it. This puts a 3-trillion-parameter open-weight model with massive context into direct competition with closed models like GPT-4 and Claude. The open-weight release lets developers self-host and fine-tune a model that rivals proprietary APIs in scale.