Moonshot AI releases Kimi K3.1 Pro with lower API costs
The new model ranks fourth on intelligence benchmarks and cuts inference prices by 33 percent.
By Alina Carter ·
Moonshot AI has released Kimi K3.1 Pro, a reasoning model that scores 55 points on the Artificial Analysis Intelligence Index.
The new version ranks fourth among all models, placing it just three points behind the top proprietary systems. Only Anthropic's Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full score higher.
Three main factors define the launch of K3.1 Pro.
First, the model shows major gains in agentic tasks. On the GDPval-AA benchmark, K3.1 Pro reached an Elo rating of 1,846. This is 312 points higher than its predecessor, Kimi K3. It trails only Claude Opus 5.5 at 1,866 and Muse Wonder Full at 1,857.
Second, inference costs have dropped sharply. API input prices are now $2 per million tokens, down from previous levels. Output costs sit at $10 per million tokens, and cache hits cost $0.20. This represents a 33 percent reduction in API prices.
Third, the company is preparing to release the model weights in an open format. Moonshot plans to publish these weights by November 4, 2026. If this happens, K3.1 Pro would become the highest-scoring open-weights model available. That move would close the performance gap between open and proprietary models to three points.
The model architecture remains consistent with the previous K3 version. It features 2.8 trillion total parameters, with 104 billion active. It supports a one-million-token context window and handles both text and image inputs. Artificial Analysis places the model on the Pareto frontier, indicating no comparable-sized model scores higher on its index.
Performance in specific technical areas also improved. On Terminal-Bench 4.0, the model achieved a 65.7 percent score. This beats GPT-6 Astra, which scored 58.2 percent, and sits slightly below Claude Opus 5.5 at 66.4 percent. The previous Kimi K3 model scored only 13 percent on this specific test.
Moonshot reports a decline in hallucinations during internal testing. Claim-level hallucinations dropped from 23.4 percent with K3 to 8.3 percent with K3.1 Pro on the Kimi Hallu Bench Hard. The company notes that GPT-6 Astra still leads this internal metric at 5.5 percent.
For the open-weights release, Moonshot removed cybersecurity-related data from both training and reinforcement learning phases. In internal tests, the uncleaned model solved 90.5 percent of vulnerability tasks, while the cleaned version solved 81.0 percent. The company aims to gradually expand the capabilities of open models for defensive work while managing offensive potential.
The release includes a variant called K3.1 Pro Highspeed, which the company says runs two to three times faster. Both versions are accessible through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. The model also supports 3D modeling workflows for game engines and CAD software without requiring external connectors.
