Moonshot AI's Kimi K3.1 Pro Narrows Gap With Anthropic's Claude Opus 5.5
Moonshot AI's newest reasoning model is now just three points behind the leading proprietary models on a major intelligence benchmark, with plans to release its weights soon.
By Tariq Mahmoud ·
A three-point difference. That is the gap between Chinese AI company Moonshot AI's new reasoning model, Kimi K3.1 Pro, and the top-scoring proprietary models on the Artificial Analysis Intelligence Index. The model's score of 55 places it fourth, just behind Claude Opus 5.5 at 58, and tying Claude Sonnet 5.5 and Muse Wonder Full at 56. This development matters because it signifies a rapid narrowing of the performance gap between proprietary models, which are closed systems, and open-weights models, whose underlying architecture is made available for public use.
The main claim from Moonshot AI is that this performance increase comes with no increase in model size. Kimi K3.1 Pro maintains the same dimensions as its predecessor, K3, with 2.8 trillion total parameters and 104 billion active ones. Despite this, the model achieves an 11-point jump over K3 on the Intelligence Index. The company has also reduced the cost to access this capability through its API, lowering prices for tokens by 33 percent. This combination of higher performance and lower cost is the central argument for the model's significance.
The most significant performance gains appear in agentic tasks, which involve the model's ability to execute multi-step processes. On the GDPval-AA benchmark, the model's score of 1,846 Elo is 312 points higher than its predecessor, placing it just behind Claude Opus 5.5 and Muse Wonder Full. Similarly, on the Terminal-Bench 4.0 test, K3.1 Pro scores 65.7 percent, a substantial leap from the 13 percent scored by K3. This figure is just behind Claude Opus 5.5's 66.4 percent and surpasses GPT-6 Astra's 58.2 percent.
Another area of marked improvement is the reduction of hallucinations, where a model presents false information as fact. Moonshot AI reports that its internal Kimi Hallu Bench Hard, a stress test for hallucination, shows claim-level hallucinations dropping from 23.4 percent with K3 to 8.3 percent with K3.1 Pro. The company explicitly notes this is a stress test and not a general measure of error rates. Even with this improvement, GPT-6 Astra performs better on the same benchmark with a 5.5 percent rate.
The model's capabilities extend to specialized fields. On internal finance benchmarks, K3.1 Pro's performance improved by 7.2 points on Financial Modeling and 11.6 points on Financial Research. It also tops all tested models on the OfficeQA Pro v2 benchmark for general office tasks. For 3D modeling, the model can interact with game engines like Unity, Unreal Engine, and Godot, and on the BenchCAD benchmark, it scores 92.7 percent with tools, nearing the performance of Claude Opus 5.5 (96.2 percent) and GPT-6 Astra (95.9 percent).
Moonshot AI has announced that it plans to release the model weights by November 4, 2026. If this release proceeds, K3.1 Pro would become the highest-scoring open-weights model available. This would reduce the performance gap between open and proprietary models to three points, assuming no other models advance in the interim.
Several open questions remain. The timeline for the open-weights release is a key factor in whether the model will actually become the leading open-weights option. Additionally, the model's performance on offensive cybersecurity capabilities has been deliberately limited for the open release. To prepare for this, Moonshot AI removed cyber-related data from its training phases, a process that lowered its score on an internal vulnerability research test from 90.5 percent to 81.0 percent. This is still a significant increase over the 33.3 percent scored by K3, though it remains below DeepSeek-V4.1-Flash's 85.7 percent.
Moonshot AI's stated position is that no open-weights model should push the absolute frontier of offensive cyber capabilities, while acknowledging that such models serve defensive purposes as well. This highlights a broader tension in the industry regarding the distribution of powerful AI technology. The release of K3.1 Pro's weights will be a moment to watch, as it would directly impact the competitive landscape for open-weights models.
Source: The Decoder
