AI Race: Moonshot's New Model Narrows the Gap

Moonshot AI has released Kimi K3.1 Pro, a move that puts it within striking distance of the industry leader. The new model landed in fourth place on the Artificial Analysis Intelligence Index with a score of 55. It sits just three points behind Claude Opus 5.5, which holds the top spot with 58 points. This result marks a significant jump for the Chinese company. The score is an 11 point increase over its predecessor, Kimi K3. The model now ranks ahead of several other major competitors. Only Claude Sonnet 5.5 and Muse Wonder Full score higher than the new entry. This positions Kimi K3.1 Pro as a serious contender in the current landscape.

The technical specifications remain consistent with the previous version. The model retains a total of 2.8 trillion parameters. It uses 104 billion active parameters during operation. The context window spans one million tokens. The system handles both text and image inputs. A key change is the pricing structure. API costs have dropped by 33 percent. Input tokens now cost $2 per million, down from $3. Output tokens are priced at $10, reduced from $15. Cache hits cost $0.20, a decrease from $0.30. Moonshot plans to release the model weights by November 4, 2026. This release would make it the highest scoring open weights model available. The gap between open and proprietary models would shrink from 12 points to just 3.

Performance improvements are most visible in agentic tasks. The model achieves an Elo score of 1,846 on the GDPval-AA benchmark. This is 312 points higher than the previous version. On Terminal-Bench 4.0, it records a success rate of 65.7 percent. This figure surpasses GPT-6 Astra, which manages 58.2 percent. It falls slightly short of Claude Opus 5.5, which achieves 66.4 percent. The previous Kimi K3 model scored only 13 percent on that specific test. Hallucination rates have also declined significantly. Internal tests show claim level hallucinations fell from 23.4 percent to 8.3 percent. While GPT-6 Astra still holds an edge at 5.5 percent, the reduction is substantial for Moonshot.

The model demonstrates strong capabilities in coding and 3D modeling. It works 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 Claude Opus 5.5, which scores 96.2 percent. Without tools, the model hits 72.8 percent. This nearly matches the 73.0 percent achieved by Claude Opus 5.5. Moonshot notes the model excels at parametric modeling of mechanical parts. It can write and run scripts for Blender and CAD software. The company has also launched a feature called 3D Make on its website. This allows users to preview, edit, and export 3D models directly.

Moonshot has addressed cybersecurity concerns in its technical documentation. The company observed that vulnerability research capabilities grew as post training scaled. Cyber related content was embedded in the general training corpus. For the open weights release, the team stripped this data from both training phases. On an internal benchmark, the uncleaned version solved 90.5 percent of tasks. After data cleaning, K3.1 Pro solves 81.0 percent. DeepSeek V4.1 Flash reaches 85.7 percent on the same test. The previous Kimi K3 model scored only 33.3 percent. Moonshot believes open models should not push the absolute frontier of offensive cyber capabilities. However, the company wants to gradually raise their defensive utility. This includes auditing legacy systems. The model is available now through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. A variant called K3.1 Pro Highspeed runs two to three times faster.

The release marks a pivotal moment in the competition between Chinese and Western AI firms. The narrowing score gap suggests that open models are becoming more competitive. The lower pricing structure makes advanced AI more accessible to developers. The upcoming release of model weights will further democratize access. Users can now test the model across various platforms. The improvements in agentic tasks and coding are particularly noteworthy. The reduction in hallucinations addresses a common pain point for users. The cybersecurity measures show a deliberate approach to balancing capability and safety. The market will watch closely as the open weights version arrives in November. This release could reshape the balance of power in the AI sector. The focus on practical applications like 3D modeling highlights the model's versatility.