Moonshot's Kimi K3.1 Pro Narrows Gap With Top AI Models

A new artificial intelligence model from China is rapidly closing the performance gap with leading Western competitors.

Moonshot AI has introduced Kimi K3.1 Pro, a post-trained iteration of its previous system that has surged to the fourth position on the Artificial Analysis Intelligence Index. The model achieved a score of 55 points, representing an 11-point improvement over its predecessor, Kimi K3. This places it just three points behind the current leaders, Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full. If Moonshot follows through on its plan to release the model weights by November 4, 2026, K3.1 Pro would become the highest-scoring open-weights model available. This development would reduce the performance disparity between open and proprietary systems from 12 points to just three, assuming no other major releases occur in the interim.

The most significant advancements in K3.1 Pro are evident in agentic tasks, where the model demonstrates substantial capability gains. On the GDPval-AA benchmark, the system reached an Elo rating of 1,846, marking a 312-point increase over the previous version. This score places it slightly behind Claude Opus 5.5 and Muse Wonder Full. In coding evaluations, specifically on Terminal-Bench 4.0, Moonshot reported a score of 65.7 percent, surpassing GPT-6 Astra and trailing only Claude Opus 5.5 by a narrow margin. The previous iteration, Kimi K3, had managed only 13 percent on this same test. Additionally, internal testing indicates a sharp reduction in hallucination rates, with claim-level errors dropping from 23.4 percent to 8.3 percent, although GPT-6 Astra maintains a lower error rate of 5.5 percent on the same internal benchmark.

The model also exhibits strong capabilities in 3D modeling and general coding applications. Moonshot states that K3.1 Pro can interact directly with game engines such as Unity, Unreal Engine, and Godot, handling everything from minor adjustments to complete project builds. On the BenchCAD benchmark, the model achieved 92.7 percent accuracy when using tools, a result close to the scores of Claude Opus 5.5 and GPT-6 Astra. Without external tools, it scored 72.8 percent, nearly matching the performance of Claude Opus 5.5. The company highlighted its strength in parametric modeling for mechanical parts and architecture, noting that the model can write and execute scripts for software like Blender and CAD tools without requiring specific connectors. A new feature called "3D Make" has also been added to allow users to preview and export models.

Despite the performance upgrades, the underlying architecture remains consistent with the previous version, retaining 2.8 trillion total parameters and 104 billion active parameters. The model supports a one-million-token context window and handles both text and image inputs. According to Artificial Analysis, it sits at the top of the Pareto frontier for its size class. However, the cost of using the API has decreased significantly. Input prices have dropped to $2 per million tokens, down from $3, while output costs have fallen to $10 per million tokens from $15. Cache hit prices have also been reduced to $0.20. These price cuts mean the cost per Intelligence Index task remains nearly identical to the previous version, hovering around $1.98, as the model generates slightly more output tokens.

Moonshot has also addressed cybersecurity considerations in its technical documentation. The company noted that vulnerability research capabilities grew as post-training expanded, leading them to remove cyber-related data from the training process for the open-weights release. On an internal test that reproduces recently disclosed vulnerabilities in software like the Linux kernel, the uncleaned version of the model solved 90.5 percent of tasks. After data cleaning, this figure dropped to 81.0 percent, still outperforming the previous Kimi K3 score of 33.3 percent. The company aims to gradually increase the offensive capabilities of open models for defensive purposes, while ensuring they do not push the absolute frontier of cyberattacks.

The new model is available immediately through various platforms, including Kimi.ai, Kimi Work, and the Kimi API. Moonshot also released a variant called K3.1 Pro Highspeed, which the company says runs two to three times faster than the standard version. For New York and New Jersey tech professionals, this release signals a continued intensification of the global AI race, where cost efficiency and open access are becoming critical competitive factors. The upcoming release of weights in November could further shift the balance of power in the industry, potentially lowering barriers to entry for developers who rely on open-source tools.

The competitive landscape remains dynamic, with proprietary models still holding a slight edge in raw benchmark scores. However, the shrinking gap suggests that the distinction between open and closed systems is becoming less pronounced. As Moonshot prepares to share its model weights, other developers may look to replicate these gains, potentially accelerating innovation across the sector. The focus on agentic tasks and 3D modeling indicates a broader shift toward AI systems that can execute complex, multi-step workflows rather than simply generating text. This trend is likely to influence how businesses in the region integrate AI into their daily operations over the coming months.

Analysts watching the AI market will closely monitor the November release of the model weights. If K3.1 Pro maintains its high scores once widely adopted, it could challenge the dominance of established proprietary systems. The price reductions also make high-performance AI more accessible to smaller firms and startups, which could drive a wave of new applications in software development and design. The coming weeks will reveal how quickly other competitors respond to these changes, potentially triggering further price cuts or performance upgrades across the board. The race to the top remains intense, with each new release setting the stage for the next round of competition.