Kimi K3.1 Pro Hits 55 Points, Cutting AI Price Gap to 3% of Leaders

Moonshot AI’s new reasoning model, Kimi K3.1 Pro, has secured a score of 55 on the Artificial Analysis Intelligence Index, placing it just three points behind the market-leading proprietary systems. This performance leap positions the Chinese firm’s latest release as a significant competitive threat in the enterprise AI sector, challenging the dominance of established Western players.

The model’s primary value proposition lies in its balance of high performance and reduced operational costs. While it trails the top proprietary models by a narrow margin, it delivers this capability at a 33% lower price point per token. This cost efficiency is critical for businesses scaling their AI workloads, as it allows for higher volume processing without a proportional increase in infrastructure spend.

Market positioning for K3.1 Pro is set to strengthen further if the company adheres to its release schedule. Moonshot plans to publish the model’s weights by November 4, 2026. If executed on time, K3.1 Pro would become the highest-scoring open-weights model available. This milestone would effectively shrink the performance gap between open-source and proprietary alternatives from 12 points to just three, assuming no other major releases intervene.

The technical gains are most pronounced in agentic tasks and specialized coding applications. On the GDPval-AA benchmark, the model achieved an Elo rating of 1,846, a 312-point increase over its predecessor. It also posted a 65.7% score on Terminal-Bench 4.0, surpassing GPT-6 Astra and arriving within less than one percentage point of Claude Opus 5.5. These results suggest the model is well-suited for autonomous workflow automation and complex software development environments.

Cost structures have been adjusted to reflect the improved efficiency of the new architecture. Input costs now stand at $2 per million tokens, down from $3, while output costs have dropped to $10 from $15. Cache hit pricing also fell to $0.20. Although the model generates 14% more output tokens than its predecessor, the total cost per Intelligence Index task remains nearly identical at $1.98, maintaining budgetary predictability for developers.

The model maintains the same architectural footprint as its predecessor, featuring 2.8 trillion total parameters with 104 billion active ones. It supports a one-million-token context window and multimodal inputs including text and images. According to Artificial Analysis, K3.1 Pro sits at the top of the Pareto frontier, meaning no model with a comparable parameter count currently scores higher on the intelligence metrics.

A key differentiator for the upcoming open-weights release is the deliberate limitation of offensive cybersecurity capabilities. Moonshot removed cyber-related data from the training corpus to manage risk. On internal benchmarks, the uncleaned version solved 90.5% of vulnerability tasks, whereas the cleaned K3.1 Pro solved 81.0%. This strategic choice aims to balance defensive utility with safety, acknowledging that open models serve both offensive and defensive roles in the security landscape.

Moonshot is currently offering K3.1 Pro through its standard API and web interfaces, alongside a high-speed variant that runs two to three times faster. The company also introduced a 3D modeling feature that allows users to preview and export models directly. With the weights release imminent, the market may see a shift in how developers weigh the trade-offs between proprietary convenience and open-source flexibility.