Kimi K3.1 Pro Cuts AI Cost Gap by 33% While Rivaling Top Proprietary Models
Moonshot AI's new model achieves near-top benchmark scores with lower pricing and plans to release open weights by November.
By Kaito Sokolov ·
A 33 percent reduction in API pricing is the headline number behind the launch of Kimi K3.1 Pro, a new reasoning model from Chinese AI developer Moonshot AI. This price cut accompanies a significant performance leap, positioning the model as a direct competitor to the leading proprietary systems currently dominating the market. The release marks a notable shift in the balance of power between closed-source giants and emerging open-weight contenders.
The most immediate impact is found in the benchmark rankings. Kimi K3.1 Pro has secured the fourth spot on the Artificial Analysis Intelligence Index with a score of 55 points. This represents an 11-point improvement over its predecessor, Kimi K3. Currently, only three models rank higher: Claude Opus 5.5 with 58 points, followed by Claude Sonnet 5.5 and Muse Wonder Full, both at 56 points. This places the new model within striking distance of the industry leaders.
Performance gains are most evident in agentic tasks, where the model handles complex, multi-step workflows. On the GDPval-AA benchmark, K3.1 Pro achieves an Elo rating of 1,846, surpassing its predecessor by 312 points. It trails only Claude Opus 5.5 and Muse Wonder Full in this specific metric. Additionally, on Terminal-Bench 4.0, the model records a success rate of 65.7 percent, significantly outperforming GPT-6 Astra at 58.2 percent and closely matching the top performer, Claude Opus 5.5, at 66.4 percent.
The model also demonstrates improved reliability and specialized capabilities. Hallucination rates have dropped sharply, with claim-level errors falling from 23.4 percent in the previous version to 8.3 percent in K3.1 Pro, according to internal testing. While GPT-6 Astra maintains a slight edge with a 5.5 percent error rate on the same stress test, Moonshot notes that the new model excels in financial modeling and research, gaining 7.2 and 11.6 points respectively. For general office tasks, the company reports it tops all tested models on the OfficeQA Pro v2 benchmark.
In the realm of creative and technical construction, K3.1 Pro integrates directly with major game engines such as Unity, Unreal Engine, and Godot. It scores 92.7 percent on the BenchCAD benchmark when using tools, a figure close to the 96.2 percent achieved by Claude Opus 5.5. Without external tools, it still manages 72.8 percent, nearly matching the top proprietary model. The company has also introduced a feature called "3D Make" on its platform, allowing users to preview, edit, and export 3D models, with the model showing particular strength in parametric modeling for mechanical and architectural design.
Crucially, the architectural size of the model remains unchanged. It retains the 2.8 trillion total parameters and 104 billion active parameters of its predecessor, while supporting a one-million-token context window for both text and image inputs. According to Artificial Analysis, this efficiency places it at the top of the Pareto frontier, meaning no other model with a similar parameter count achieves a higher Intelligence Index score. The reduced token prices mean that while the model generates 14 percent more output tokens, the cost per task remains nearly identical to the previous version at $1.98.
Moonshot AI is also addressing cybersecurity concerns ahead of its planned open-weights release. The company states that vulnerability research capabilities increased naturally as post-training scaled, due to cyber-related content in the training data. To manage this, Moonshot removed specific cyber data from both training phases for the open version. On an internal benchmark reproducing recent software vulnerabilities, the uncleaned version solved 90.5 percent of tasks, while the cleaned K3.1 Pro version solved 81.0 percent. This is higher than DeepSeek-V4.1-Flash at 85.7 percent, but the company aims to balance offensive capability with defensive utility.
If released as scheduled by November 4, 2026, K3.1 Pro would become the highest-scoring open-weights model available. This move would shrink the performance gap between open and proprietary models from 12 points to just three, assuming no other major releases occur in the interim. The company has made both the standard and a "Highspeed" variant, which runs two to three times faster, available now through its various platforms and APIs.
The broader context suggests a tightening race where cost efficiency and open access are becoming critical differentiators. By maintaining the same hardware requirements while lowering costs and improving accuracy in key areas like coding and 3D modeling, Moonshot AI is challenging the premium pricing structure of established leaders. The upcoming release of model weights will allow developers to fine-tune the system, potentially further narrowing the technological divide in the AI landscape.
