Moonshot AI Narrows Gap With Top Rivals Via Kimi K3.1 Pro

Kimi K3.1 Pro has just entered the conversation about top-tier artificial intelligence, securing the fourth position on the Artificial Analysis Intelligence Index. The model achieves a score of 55, placing it only three points behind the current leader, Anthropic's Claude Opus 5.5. This rapid ascent marks a significant shift in the competitive landscape for large language models.

The release represents a post-trained evolution of the earlier Kimi K3, with Moonshot AI applying substantially more computational resources during the final training phase. This investment resulted in an 11-point improvement over its predecessor. Currently, only Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full hold higher rankings. If Moonshot releases the model weights by the planned date of November 4, 2026, Kimi K3.1 Pro would become the highest-scoring open-weights model available. Such a move would compress the performance gap between open and proprietary systems from 12 points down to just 3, assuming no other major releases occur in the interim.

The most notable gains in K3.1 Pro appear in agentic tasks, where the model demonstrates increased reliability. On the GDPval-AA benchmark, it reaches an Elo rating of 1,846, a jump of 312 points from K3. This places it narrowly behind Claude Opus 5.5 and Muse Wonder Full. In terminal-based coding tests, the model reports a success rate of 65.7 percent on Terminal-Bench 4.0. This figure exceeds GPT-6 Astra's 58.2 percent and sits only slightly below Claude Opus 5.5's 66.4 percent, whereas K3 managed just 13 percent on the same evaluation.

Accuracy has also improved, with internal testing showing a sharp decline in hallucinations. Moonshot reports that claim-level hallucinations dropped from 23.4 percent in K3 to 8.3 percent in the new version. While the company notes this is a stress test rather than a measure of general error rates, GPT-6 Astra still holds an advantage with a 5.5 percent rate on the same benchmark. Financial modeling and research capabilities also saw gains of 7.2 and 11.6 points respectively, and the model reportedly tops all tested competitors on the OfficeQA Pro v2 general office task suite.

Specialized capabilities in three-dimensional modeling and coding have expanded significantly. The model now interacts directly with game engines such as Unity, Unreal Engine, and Godot, handling everything from quick fixes to full project builds. On the BenchCAD benchmark, it scores 92.7 percent when using tools, close to the scores of Claude Opus 5.5 and GPT-6 Astra. Without tools, it achieves 72.8 percent, nearly matching Claude Opus 5.5. Moonshot highlights its strength in parametric modeling of mechanical and architectural parts, noting that the model can write and run scripts for Blender and CAD software without requiring external connectors.

Pricing and structural specifications remain a point of interest for developers. K3.1 Pro maintains the same 2.8 trillion total parameters and 104 billion active parameters as its predecessor. It supports a one-million-token context window and processes both text and images. According to Artificial Analysis, it sits at the top of the Pareto frontier, meaning no model with a comparable size scores higher. Token prices have dropped by 33 percent, with input costs now at $2 per million tokens and output costs at $10. Cache hits are priced at $0.20. The cost per Intelligence Index task is $1.98, nearly identical to the previous version due to a 14 percent increase in output tokens.

Moonshot has addressed cybersecurity considerations in its technical documentation, noting that vulnerability research capabilities grew as post-training scaled. For the upcoming open-weights release, the company stripped cyber-related data from both training phases. On an internal benchmark reproducing recently disclosed vulnerabilities, the uncleaned version solved 90.5 percent of tasks, while the cleaned K3.1 Pro solved 81.0 percent. DeepSeek-V4.1-Flash scored 85.7 percent, and the older Kimi K3 scored 33.3 percent. The company argues that open-weights models should not push the absolute frontier of offensive cyber capabilities but should gradually improve to support defensive tasks like auditing legacy systems.

Both K3.1 Pro and its faster variant, K3.1 Pro Highspeed, are available now through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. The Highspeed version runs two to three times faster, offering a trade-off between speed and depth. This release signals a continued push by Chinese developers to close the performance gap with leading Western models while making advanced capabilities more accessible through lower prices and open weights.