Moonshot AI Cuts Costs on New Kimi K3.1 Pro Model

The competitive landscape for large language models is tightening as Chinese developers push for cost-efficiency. Moonshot AI has officially launched Kimi K3.1 Pro, a post-trained iteration of its predecessor that has immediately reshaped the market rankings.

The new model secured fourth place on the Artificial Analysis Intelligence Index with a score of 55 points. This represents an 11-point improvement over the previous K3 version, placing it within three points of the top proprietary offerings. Only Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full currently rank higher, making the margin for error negligible in the current hierarchy.

A critical factor for enterprise adoption is the reduction in operational costs. While the model retains the same architectural scale of 2.8 trillion total parameters with 104 billion active, API pricing has dropped by 33 percent. Input tokens now cost $2 per million, down from $3, while output tokens are priced at $10, a decrease from $15. This price cut ensures that the cost per Intelligence Index task remains nearly identical to the previous model, despite the new version generating 14 percent more output tokens.

Performance metrics indicate significant advancements in specialized tasks, particularly in agentic workflows and coding. On the GDPval-AA benchmark, the model achieved an Elo rating of 1,846, a substantial leap from the previous version. It also outperformed GPT-6 Astra on Terminal-Bench 4.0, recording a 65.7 percent success rate. Additionally, the model demonstrates superior capabilities in 3D modeling, supporting direct integration with game engines like Unity and Unreal Engine, and scoring 92.7 percent on the BenchCAD test with tools enabled.

Moonshot plans to release the model weights by November 4, 2026. If executed on schedule, Kimi K3.1 Pro would become the highest-scoring open-weights model available. This move would shrink the performance gap between open and proprietary systems from 12 points to just three, fundamentally altering the value proposition for organizations seeking to own their infrastructure. The company also introduced a high-speed variant that runs two to three times faster, available immediately through its various API and web platforms.

The release highlights a strategic balance between offensive capability and stability. Moonshot reports that claim-level hallucinations dropped from 23.4 percent to 8.3 percent on internal stress tests. However, the company applied specific data cleaning for the open-weights release to manage cybersecurity capabilities. While the uncleaned version solved 90.5 percent of tasks in vulnerability reproduction, the cleaned model solves 81.0 percent. This approach aims to maintain defensive utility for auditing legacy systems without pushing the absolute frontier of offensive cyber capabilities, a concern previously highlighted by Anthropic regarding other open models.