Chinese AI Model Narrows Gap With Leaders

Moonshot AI’s new Kimi K3.1 Pro model has surged to fourth place on the Artificial Analysis Intelligence Index, placing it within striking distance of the top proprietary systems. This post-trained update represents a significant leap forward for the Chinese firm, which aims to close the widening gap between open-weights and closed-source artificial intelligence. By securing a score of 55 points, the model sits just three points behind the leading contenders, marking a dramatic shift in the competitive landscape for large language models.

The performance gains are substantial. The model recorded an 11-point increase over its predecessor, Kimi K3, to achieve its current standing. According to Artificial Analysis, only three models currently rank higher: Anthropic’s Claude Opus 5.5 with 58 points, followed by Claude Sonnet 5.5 and Muse Wonder Full, both with 56 points. If Moonshot releases the model weights as scheduled by November 4, 2026, K3.1 Pro would become the highest-scoring open-weights option available. This move would shrink the disparity between open and proprietary models from 12 points to a mere three, assuming no other major releases occur in the interim.

The improvements are most visible in agentic tasks and coding capabilities. On the GDPval-AA benchmark, K3.1 Pro reaches 1,846 Elo, a 312-point jump from the previous version. It trails only Claude Opus 5.5 and Muse Wonder Full in this specific metric. In terminal-based tasks, the model achieves 65.7 percent on Terminal-Bench 4.0, surpassing GPT-6 Astra and closely following Claude Opus 5.5. The predecessor model managed only 13 percent on that same test, highlighting the scale of the recent upgrade. Additionally, internal stress tests show a sharp reduction in hallucinations, with claim-level errors dropping from 23.4 percent to 8.3 percent, though GPT-6 Astra still holds a slightly lower error rate.

Beyond text, the model excels in 3D modeling and direct integration with game engines like Unity and Unreal. It scores 92.7 percent on the BenchCAD benchmark when using tools, approaching the performance of top-tier competitors. Moonshot also launched a feature called "3D Make" on its platform, allowing users to preview and edit 3D models directly. The company notes that the model is particularly strong in parametric modeling for mechanical parts and architecture, capable of writing and running scripts for CAD software without additional connectors.

Despite the performance boost, the underlying architecture remains the same size as K3, with 2.8 trillion total parameters and 104 billion active ones. The model supports a one-million-token context window and processes both text and images. However, the cost to use the model has dropped by 33 percent. Input pricing is now $2 per million tokens, down from $3, while output costs have fallen to $10 from $15. A high-speed variant, K3.1 Pro Highspeed, is also available, offering two to three times faster processing speeds. Both versions are accessible through Kimi.ai and related professional tools.

A notable aspect of the release is Moonshot’s approach to cybersecurity capabilities. The company removed cyber-related data from the training corpus for the open-weights version to prevent the model from becoming too powerful in offensive tasks. On internal benchmarks reproducing known vulnerabilities, the uncleaned model solved 90.5 percent of tasks, while the cleaned version solved 81.0 percent. This deliberate limitation aims to balance the model’s utility for defensive auditing with concerns about offensive potential, a nuanced strategy in the evolving AI safety landscape.