55 Points: Kimi K3.1 Pro Narrows AI Gap

For everyday users and businesses relying on artificial intelligence, the cost and capability of available models are shifting. A new release from Chinese developer Moonshot AI, named Kimi K3.1 Pro, has entered the market with a 33 percent reduction in pricing while simultaneously improving its performance scores. This move places the model just behind the leading proprietary options, offering a potentially more affordable alternative for tasks ranging from coding to complex office work.

The model has climbed to the fourth position on the Artificial Analysis Intelligence Index with a score of 55. This represents an 11-point increase over its predecessor, Kimi K3. Currently, only three models rank higher: Claude Opus 5.5 at 58, Claude Sonnet 5.5 at 56, and Muse Wonder Full at 56. If Moonshot releases the model's weights as planned by November 4, 2026, it would become the top-scoring open-weights model available. This would reduce the performance gap between open and proprietary models from 12 points to just 3, assuming no other major changes occur by that date.

The most significant improvements are visible in agentic tasks, where the AI performs multi-step operations autonomously. On the GDPval-AA benchmark, Kimi K3.1 Pro achieved an Elo rating of 1,846, a jump of 312 points from the previous version. This score places it slightly behind Claude Opus 5.5 and Muse Wonder Full. In terminal-based testing on Terminal-Bench 4.0, the new model reached 65.7 percent, surpassing GPT-6 Astra and coming close to the top score of 66.4 percent held by Claude Opus 5.5. The previous Kimi K3 model had scored only 13 percent on this same test.

Reliability has also improved, with a sharp decline in hallucination rates. According to Moonshot's internal Kimi Hallu Bench Hard, claim-level hallucinations dropped from 23.4 percent in K3 to 8.3 percent in K3.1 Pro. While GPT-6 Astra maintains a lower rate of 5.5 percent on this specific stress test, Moonshot notes that the reduction indicates a more stable model for practical use. The company also reports gains in financial modeling and research, as well as top scores on office task benchmarks.

Capabilities in 3D modeling and coding have expanded considerably. The model can now interact directly with game engines such as Unity, Unreal Engine, and Godot to handle project builds and fixes. On the BenchCAD benchmark, it scored 92.7 percent when using tools, approaching the results of Claude Opus 5.5 and GPT-6 Astra. Without external tools, it achieved 72.8 percent, nearly matching the top proprietary model. Moonshot states the model is particularly strong in parametric modeling for mechanical parts and architecture, and it can write scripts for Blender and CAD software without additional connectors.

Despite the performance boost, the model retains the same architecture as its predecessor, with 2.8 trillion total parameters and 104 billion active ones. It supports a context window of one million tokens and processes both text and images. According to Artificial Analysis, it sits at the top of the Pareto frontier, meaning no other model with a similar parameter count scores higher on the Intelligence Index. A faster variant, K3.1 Pro Highspeed, is also available, running two to three times quicker than the standard version.

Pricing has been adjusted to reflect the efficiency gains. Input costs are now $2 per million tokens, down from $3, while output costs dropped to $10 from $15. Cache hits are priced at $0.20, reduced from $0.30. Although the model generates 14 percent more output tokens, the cost per Intelligence Index task remains nearly identical to the previous version at $1.98.

Moonshot has also addressed cybersecurity concerns in its technical documentation. The company observed that vulnerability research capabilities increased as post-training scaled up. For the upcoming open-weights release, cyber-related data was removed from the training phases to manage offensive capabilities. In internal testing, the uncleaned version solved 90.5 percent of vulnerability tasks, while the cleaned version solved 81.0 percent. This is comparable to DeepSeek-V4.1-Flash at 85.7 percent, but significantly higher than the previous Kimi K3 at 33.3 percent.

The company aims to gradually raise the capabilities of open-weights models for defensive purposes, such as auditing legacy systems, while keeping them from pushing the absolute frontier of offensive cyber tasks. All new models and variants are currently available through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.

The next milestone for Kimi K3.1 Pro is the release of its model weights by November 4, 2026. This release is expected to solidify its position as the leading open-weights model and further compress the performance gap with proprietary leaders. Users can access the new features, including the "3D Make" tool on Kimi.com, immediately.