Chinese AI Model Kimi K3.1 Pro Narrows Gap With Top US Rivals
Moonshot AI's latest release scores 55 points on key benchmarks, placing it just behind leading US models while cutting costs by a third.
By Isabella Morales ·
Kimi K3.1 Pro has moved to fourth place on the Artificial Analysis Intelligence Index. The new model from Moonshot AI scores 55 points. This puts it within striking distance of the top proprietary models. Only three systems currently rank higher. Claude Opus 5.5 leads with 58 points. Claude Sonnet 5.5 and Muse Wonder Full both sit at 56 points. This marks a significant leap for the Chinese company in the global AI race.
The release follows the predecessor Kimi K3. Moonshot applied significantly more post-training compute to the new version. That effort resulted in an 11-point jump in the overall intelligence score. If Moonshot releases the model weights by November 4, 2026, as planned, the gap between open and proprietary models will shrink dramatically. The difference would drop from 12 points to just 3 points. K3.1 Pro would become the highest-scoring open-weights model available.
Performance gains are most visible in agentic tasks. On the GDPval-AA benchmark, the model reaches 1,846 Elo. This is 312 points higher than K3. It trails only Claude Opus 5.5 and Muse Wonder Full. In coding, the results are equally impressive. Moonshot reports a 65.7 percent score on Terminal-Bench 4.0. That beats GPT-6 Astra, which scored 58.2 percent. It is nearly on par with Claude Opus 5.5, which achieved 66.4 percent. The older K3 model managed only 13 percent on that same test.
Hallucination rates have also improved sharply. Internal testing shows a drop in claim-level errors. The rate fell from 23.4 percent in K3 to 8.3 percent in K3.1 Pro. Moonshot notes this is a stress test, not a general measure. GPT-6 Astra still holds an edge with a 5.5 percent error rate. The new model also performs better in finance. It gained 7.2 points in Financial Modeling and 11.6 in Financial Research. It tops all tested models on OfficeQA Pro v2 for general office tasks.
3D modeling capabilities are a major highlight. The model works directly with game engines like Unity, Unreal Engine, and Godot. It can handle quick fixes or build full projects. On the BenchCAD benchmark, it scores 92.7 percent with tools. This is close to Claude Opus 5.5 at 96.2 percent. Without tools, it hits 72.8 percent, nearly matching Claude Opus 5.5 at 73.0 percent. Moonshot says the model excels at parametric modeling for mechanical and architectural parts. Users can now preview and export 3D models via a new feature on Kimi.com.
The model keeps the same size as its predecessor. It has 2.8 trillion total parameters and 104 billion active ones. The context window spans one million tokens. It handles both text and images. Despite the same size, prices have dropped by 33 percent. Input costs are now $2 per million tokens. Output costs are $10. Cache hits cost $0.20. The cost per Intelligence Index task is $1.98. This is nearly identical to the previous model, which was $2.00.
Moonshot addressed cybersecurity in its technical blog. The company observed that vulnerability research capabilities grew as post-training scaled. To prepare for the open-weights release, they stripped cyber-related data from training. On an internal benchmark, the uncleaned version solved 90.5 percent of tasks. After cleaning, K3.1 Pro solves 81.0 percent. DeepSeek-V4.1-Flash scores 85.7 percent on the same test. The older K3 scored only 33.3 percent.
Moonshot believes no open-weights model should push the absolute frontier of offensive cyber capabilities. However, the company wants to gradually raise what open models can do. This helps with defensive purposes like auditing legacy systems. Anthropic previously showed that even lower-scoring models can write exploits at high levels. The new 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 signals a tightening market. Chinese firms are closing the gap with US leaders. The combination of higher scores and lower prices creates pressure on competitors. The upcoming release of weights could change the landscape further. Developers will have access to a powerful open model. This may accelerate innovation in agentic and coding applications. The next few weeks will be critical. The November deadline for the weight release is near. The industry will watch closely to see if the predicted gap reduction holds.
Source: The Decoder
