China's Moonshot AI launched Kimi K3.1 Pro on October 24, 2026, positioning it as an open-weight model within striking distance of the frontier. According to the company, it is a 3-trillion-parameter-class model with a 1-million-token context window — built for the long, multi-step agent work where smaller models typically run out of steam.

The benchmark numbers are the headline. Moonshot says K3.1 Pro scores 55 on the Artificial Analysis Intelligence Index, ranking 4th overall, and 65.7% on Terminal-Bench 4.0 — ahead of GPT-6 Astra's reported 58.2% and just behind Claude Opus 5.5's 66.4% on the same test. These are vendor-reported and third-party-benchmark figures, not independent Aitopkit tests, so treat them as directional until the community reproduces them.

Pricing is set at $2 per million input tokens and $10 per million output tokens — the same headline rate as several frontier flagships — and Moonshot says it will release the full model weights by November 4, 2026. If that release lands on schedule, developers and self-hosters get a near-frontier open model they can run and fine-tune themselves, rather than only renting through an API.

For builders, the practical question is not the parameter count but whether the long context and agentic scores hold up on real workloads once the weights are public. The November 4 release date is the one to watch — open weights turn a launch announcement into something the whole ecosystem can verify.

Sources