New AI model cuts costs and boosts performance for users

For developers and businesses relying on artificial intelligence, the cost of running complex tasks just dropped. A new model released by Moonshot AI offers significant price reductions while delivering higher performance scores. This shift means that advanced AI capabilities are becoming more accessible without requiring a larger budget for API usage.

The release of Kimi K3.1 Pro marks a notable step in the ongoing competition among AI providers. The model was not built from scratch. Instead, it is a refined version of the previous Kimi K3 model. Moonshot applied significantly more post-training compute to improve its reasoning abilities. This approach allowed the company to enhance specific skills without altering the underlying structure of the software.

The new model has quickly climbed the rankings. It now sits at fourth place on the Artificial Analysis Intelligence Index with a score of 55. Only three proprietary models rank higher: Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full. The gap between this new Chinese model and the top proprietary systems is narrow, standing at just three points.

Performance gains are most visible in agentic tasks and coding. On the GDPval-AA benchmark, the model achieves an Elo rating of 1,846, which is a substantial increase from its predecessor. It also performs well on Terminal-Bench 4.0, where it records a success rate of 65.7 percent. This figure surpasses GPT-6 Astra and is nearly on par with Claude Opus 5.5. The previous version of the model managed only 13 percent on that same test, highlighting the magnitude of the improvement.

The model also shows stronger capabilities in 3D modeling and creative coding. It can interact directly with game engines such as Unity, Unreal Engine, and Godot. Moonshot reports that the model is particularly effective at parametric modeling of mechanical parts and architectural structures. Users can now write and run scripts for Blender and CAD software without needing additional connectors. The company has also added a feature on its website that allows users to preview and export 3D models directly.

Pricing has been adjusted to reflect the improved efficiency. The total parameter count remains at 2.8 trillion, with 104 billion active parameters. However, the cost per million input tokens has fallen to $2, down from $3. Output costs are now $10 per million tokens, a reduction from $15. Cache hits cost $0.20. These lower rates make the model more attractive for high-volume applications.

Security considerations have also shaped the release. Moonshot noted that vulnerability research capabilities increased during training. For the open-weights version, the company removed cyber-related data to manage offensive capabilities. On an internal test involving recent software vulnerabilities, the cleaned model solves 81 percent of tasks. This is slightly lower than the uncleaned version but still higher than previous iterations. The company aims to balance offensive potential with defensive utility for auditing legacy systems.

The next milestone is the release of the model weights. Moonshot plans to make these available by November 4, 2026. If this schedule holds, Kimi K3.1 Pro will become the highest-scoring open-weights model. That move would narrow the performance gap between open and proprietary systems to just three points. A faster variant called K3.1 Pro Highspeed is also available now, running two to three times quicker than the standard version.

This development continues to reshape the AI landscape. As costs decrease and performance rises, the distinction between closed and open models becomes less significant. Users gain more options for choosing tools that fit their specific needs and budgets. The focus on practical tasks like coding and 3D design suggests a shift toward AI that does more than just generate text.