Kimi K3.1 Pro nears top AI ranks
Moonshot AI's latest model jumps to fourth place on key benchmarks, significantly narrowing the gap with leading proprietary systems.
By Samir Nasser ·
The race for the best large language model is tightening. Chinese company Moonshot AI has released Kimi K3.1 Pro, which now sits in fourth place on the Artificial Analysis Intelligence Index with a score of 55. This result places it just behind Anthropic's Claude Opus 5.5, which holds the top spot with 58 points. Other models like Claude Sonnet 5.5 and Muse Wonder Full also rank above the new entry.
The jump to fourth place represents an 11 point improvement over the previous version, Kimi K3. The model maintains the same architecture as its predecessor, featuring 2.8 trillion total parameters and 104 billion active parameters. Despite the larger scale, the company reduced API costs by 33 percent. Input tokens now cost $2 per million, while output tokens are $10 per million. Moonshot plans to release the full model weights by early November 2026. If this schedule holds, K3.1 Pro would become the highest scoring open weights model available.
Performance gains are most visible in agentic tasks. On the GDPval AA benchmark, the model achieves 1,846 Elo, a significant rise over the prior version. It also scores 65.7 percent on Terminal Bench 4.0, surpassing GPT 6 Astra. Hallucination rates have dropped sharply as well. Internal testing shows claim level hallucinations fell from 23.4 percent to 8.3 percent. However, GPT 6 Astra still leads with a 5.5 percent error rate on that specific stress test.
The new model also excels in 3D modeling and coding. It can work directly with game engines like Unity and Unreal Engine. On the BenchCAD benchmark, it scores 92.7 percent when using tools. This is close to the top tier models. Moonshot also launched a feature called 3D Make on its website. This tool allows users to preview, edit, and export 3D models directly from the platform.
The company made specific adjustments to cybersecurity capabilities for the open weights release. Moonshot removed cyber related data from its training process to manage offensive capabilities. On an internal test for vulnerability discovery, the uncleaned version solved 90.5 percent of tasks. After data cleaning, that number dropped to 81.0 percent. The company believes this balance allows open models to serve defensive purposes without pushing the absolute frontier of offensive cyber tools.
K3.1 Pro is available now through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. A faster variant called K3.1 Pro Highspeed is also in use. It runs two to three times quicker than the standard version. The model supports a context window of one million tokens and handles both text and images. Users can access the system today at the new lower price points.
The release marks a significant shift in the open versus proprietary model landscape. By combining high performance with lower costs, Moonshot is challenging the dominance of established players. The upcoming release of model weights will allow researchers and developers to inspect the architecture in detail. This transparency may further accelerate adoption in the industry. The next few weeks will be critical as the weights become public and independent tests confirm the reported benchmarks.
Source: The Decoder
