Moonshot AI cuts Kimi API prices 33 percent as new model ranks fourth

Developers using Chinese AI provider Moonshot AI will pay less for processing data starting today. The company lowered input prices to $2 per million tokens and output prices to $10 per million tokens. These new rates are 33 percent lower than the previous pricing structure for its Kimi models.

The price reduction accompanies the launch of Kimi K3.1 Pro, a reasoning model that scored 55 points on the Artificial Analysis Intelligence Index. This result places the system in fourth position globally. It trails Anthropic's Claude Opus 5.5, which scored 58, by three points. The new model also outperformed its predecessor, Kimi K3, by 11 points.

Moonshot plans to release the model's weights by November 4, 2026. If published as scheduled, K3.1 Pro would become the highest-scoring open-weights model available. This release would narrow the performance gap between open and proprietary systems from 12 points to 3. The architecture remains consistent with the previous version, featuring 2.8 trillion total parameters and 104 billion active parameters.

Performance gains were most visible in agentic tasks. On the GDPval-AA benchmark, K3.1 Pro achieved an Elo score of 1,846. This is 312 points higher than K3 and sits just behind Claude Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. In coding, the model recorded a 65.7 percent score on Terminal-Bench 4.0. This result exceeded GPT-6 Astra's 58.2 percent and approached Claude Opus 5.5's 66.4 percent.

Moonshot also reported a decrease in hallucinations. Claim-level hallucinations dropped from 23.4 percent on K3 to 8.3 percent on the internal Kimi Hallu Bench Hard. The company noted that GPT-6 Astra still leads this specific internal test with a 5.5 percent error rate. Finance benchmarks showed similar improvements, with the new model gaining 11.6 points in financial research tasks.

The technical blog post details changes to the model's cybersecurity capabilities. Moonshot stated that vulnerability research improved as post-training scaled up. For the upcoming open-weights release, the company removed cyber-related data from the training process. On an internal benchmark involving Linux kernel and Node.js vulnerabilities, the uncleaned model solved 90.5 percent of tasks. After data removal, that figure dropped to 81.0 percent.

Both the standard and high-speed versions of K3.1 Pro are available through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. The high-speed variant operates two to three times faster than the standard model. The standard model generates 14 percent more output tokens, resulting in a cost of $1.98 per Intelligence Index task. This cost is nearly identical to the $2.00 cost for the previous K3 model.