Moonshot AI Narrows Gap With Anthropic in New Kimi Release

Moonshot AI has positioned its latest model, Kimi K3.1 Pro, within striking distance of the industry leaders, scoring 55 points on the Artificial Analysis Intelligence Index. This places the model in fourth place, just three points behind Anthropic's Claude Opus 5.5, which currently holds the top spot with a score of 58. The release marks a significant shift in the competitive landscape for large language models, as the new version represents a post-trained evolution of its predecessor, Kimi K3, built with substantially more compute resources during the training phase.

The most notable improvements in the new model are found in agentic tasks, where it demonstrates a sharp increase in capability. On the GDPval-AA benchmark, the model achieved an Elo rating of 1,846, a jump of 312 points over the previous version. It also performed well on Terminal-Bench 4.0, recording a score of 65.7 percent, which surpasses GPT-6 Astra and remains very close to Claude Opus 5.5. These gains are accompanied by a significant reduction in hallucination rates, with claim-level errors dropping from 23.4 percent to 8.3 percent on internal stress tests, although GPT-6 Astra still maintains a lower error rate of 5.5 percent.

Beyond general reasoning, the model shows specific strengths in coding and 3D modeling. It can interface directly with game engines such as Unity and Unreal Engine, and it scores 92.7 percent on the BenchCAD benchmark when using tools. Moonshot has also introduced a feature called 3D Make on its website, allowing users to preview and export models. The company notes that the model is particularly effective at parametric modeling for mechanical parts and architecture, capable of writing and running scripts for Blender and CAD software without the need for additional connectors.

Despite the performance gains, the model retains the same architectural footprint as its predecessor, featuring 2.8 trillion total parameters and 104 billion active ones. However, the cost of accessing the model has decreased by 33 percent. Input tokens now cost $2 per million, down from $3, while output tokens are priced at $10 per million, a drop from $15. The context window supports one million tokens and handles both text and image inputs. According to Artificial Analysis, the model sits at the top of the Pareto frontier, meaning no other model with a similar parameter count scores higher on the intelligence index.

The release strategy includes a plan to make the model weights open to the public by November 4, 2026. If this timeline holds, Kimi K3.1 Pro would become the highest-scoring open-weights model available, reducing the performance gap between open and proprietary systems from 12 points to just three. Moonshot also addressed cybersecurity concerns in its technical documentation, noting that it removed cyber-related data from the training corpus for the open release to balance offensive and defensive capabilities. This decision aims to allow open models to serve in defensive roles, such as auditing legacy systems, without pushing the absolute frontier of offensive cyber capabilities too far ahead of current industry standards.

The company has also made a high-speed variant available, which runs two to three times faster than the standard model. Both versions are currently accessible through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. The move signals a continued effort by Chinese AI developers to close the technical divide with US-based competitors while optimizing for cost efficiency and broader accessibility through open-weight releases.