Moonshot AI releases Kimi K3.1 Pro, cutting costs 33 percent to challenge top models
The new Chinese AI model ranks fourth on the Intelligence Index, narrowing the gap with proprietary leaders to three points while lowering API prices.
By Carolina Wood ·
Chinese artificial intelligence firm Moonshot AI released Kimi K3.1 Pro on October 23, 2026, positioning the model as a direct competitor to leading proprietary systems. The release immediately places Moonshot within striking distance of the industry leaders on key performance metrics while simultaneously reducing the price of access.
The model scored 55 points on the Artificial Analysis Intelligence Index. This result marks an 11-point improvement over its predecessor, Kimi K3. It sits just three points behind Anthropic's Claude Opus 5.5, which holds the top spot with 58 points. Claude Sonnet 5.5 and Muse Wonder Full both scored 56 points, placing them between K3.1 Pro and the leader.
Agentic tasks show the most dramatic gains. K3.1 Pro achieved 1,846 Elo on the GDPval-AA benchmark, a 312-point jump from K3. Only Claude Opus 5.5 and Muse Wonder Full scored higher on that test. On Terminal-Bench 4.0, the new model reached 65.7 percent. This figure exceeds GPT-6 Astra's 58.2 percent and trails Claude Opus 5.5 by only 0.7 points. The previous Kimi K3 model managed just 13 percent on that specific test.
Moonshot reports a sharp decline in hallucinations. On its internal Kimi Hallu Bench Hard, claim-level hallucinations dropped from 23.4 percent with K3 to 8.3 percent. The company notes that GPT-6 Astra still leads this metric at 5.5 percent. The firm describes the benchmark as a stress test for error rates rather than a general measure of model reliability.
API prices for the new model fall by 33 percent. Input tokens cost $2 per million, while output tokens cost $10. Cache hits drop to $0.20. Because the model generates 14 percent more output tokens, the cost to run each Intelligence Index task totals $1.98. This is nearly identical to the $2.00 cost for the previous K3 model.
Moonshot plans to publish the model weights by November 4, 2026. If released as planned, K3.1 Pro would become the highest-scoring open-weights model available. The gap between the best open model and top proprietary systems would shrink from 12 points to 3 points. The company is also offering a Highspeed variant that runs two to three times faster.
The architecture remains unchanged from K3, featuring 2.8 trillion total parameters with 104 billion active. The model supports a one-million-token context window for text and images. For the open-weights release, Moonshot removed cybersecurity-related data from both training phases to limit offensive capabilities. On an internal vulnerability benchmark, the uncleaned model solved 90.5 percent of tasks, while the cleaned version solved 81.0 percent.
Moonshot states that no open-weights model should push the absolute frontier of offensive cyber capabilities. The company aims to raise what open models can do gradually to support defensive work such as auditing legacy systems. The model is available through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
