Moonshot AI slashes inference costs as Kimi K3.1 Pro narrows the gap with top US rivals

Moonshot AI has introduced Kimi K3.1 Pro, a new reasoning model that has surged to fourth place on the Artificial Analysis Intelligence Index. The release marks a significant shift in the competitive landscape, positioning the Chinese developer just three points behind the leading proprietary models from Anthropic and other major players. This update arrives with a strategic focus on cost efficiency, offering developers a powerful alternative that undercuts current market pricing.

The new model scores 55 points on the Intelligence Index, representing an 11-point improvement over its predecessor, Kimi K3. It trails only Claude Opus 5.5, which holds 58 points, and Claude Sonnet 5.5 and Muse Wonder Full, both at 56 points. If Moonshot releases the model weights as planned by early November, K3.1 Pro would become the highest-scoring open-weights model available. This would reduce the performance gap between open and proprietary systems from 12 points to just three, assuming no other major updates occur before then.

Performance gains are most evident in agentic tasks and coding. On the GDPval-AA benchmark, the new model reaches 1,846 Elo, a 312-point jump from its predecessor. In terminal-based coding tests, it achieves a 65.7 percent success rate, surpassing GPT-6 Astra and falling only slightly short of Claude Opus 5.5. Moonshot also reports a sharp drop in hallucination rates, with claim-level errors falling from 23.4 percent to 8.3 percent on its internal stress test, though GPT-6 Astra still holds an edge at 5.5 percent.

Pricing has been adjusted to reflect the model's improved efficiency. While the parameter count remains at 2.8 trillion total with 104 billion active, the cost per million tokens has dropped by 33 percent. Input costs are now $2, down from $3, while output costs have fallen to $10 from $15. Cache hits are priced at $0.20, a reduction from the previous $0.30. Despite these lower unit costs, the total cost per Intelligence Index task remains roughly the same at $1.98, as the model generates 14 percent more output tokens per task.

The company is also targeting specific industries with new capabilities. K3.1 Pro supports direct interaction with game engines like Unity and Unreal, and it demonstrates strong performance in 3D modeling. On the BenchCAD benchmark, it scores 92.7 percent when using tools, approaching the results of top US competitors. Moonshot has also launched a 3D Make feature that allows users to preview and export models, highlighting the model's utility in creative and technical workflows.

A key area of focus for the upcoming open-weights release is cybersecurity. Moonshot noted that offensive capabilities grew as post-training scaled, so it removed cyber-related data from the training sets to manage risk. On an internal benchmark for reproducing vulnerabilities, the cleaned model solves 81.0 percent of tasks, down from 90.5 percent in the uncleaned version. The company aims to balance offensive potential with defensive utility, acknowledging that even smaller open models can write exploits at high levels.

K3.1 Pro is available now through various Moonshot platforms, including a high-speed variant that runs two to three times faster. The model handles both text and images within a one-million token context window. As the gap between open and proprietary models shrinks, developers may find greater flexibility in choosing between cost, speed, and performance for their applications.