Moonshot AI Launches Kimi K3.1 Pro, Undercutting Rivals on Price and Performance

The competitive landscape for large language models has shifted decisively as Chinese developer Moonshot AI introduced its latest reasoning system, Kimi K3.1 Pro. This release marks a significant escalation in the race to balance computational efficiency with advanced capabilities, directly targeting the premium segment long held by American firms.

Financially, the move is aggressive. While maintaining the same 2.8 trillion total parameter count as its predecessor, Moonshot has slashed API pricing by 33 percent. Input tokens now cost $2 per million, down from $3, while output tokens are priced at $10, reduced from $15. Cache hits have also dropped to $0.20. For enterprise buyers, this represents a substantial reduction in operational costs for high-volume inference tasks.

Performance metrics back up the pricing strategy. On the Artificial Analysis Intelligence Index, Kimi K3.1 Pro scores 55 points, an 11-point increase over the previous K3 version. This places the model in fourth position, trailing only Claude Opus 5.5 (58), Claude Sonnet 5.5 (56), and Muse Wonder Full (56). The gap between the top proprietary model and this new Chinese entrant has narrowed to just three points, signaling a tightening market.

The most pronounced gains appear in agentic workflows, a critical area for automated business processes. On the GDPval-AA benchmark, the model achieves an Elo rating of 1,846, surpassing its predecessor by 312 points. It also outperforms GPT-6 Astra on Terminal-Bench 4.0, recording a 65.7 percent success rate compared to 58.2 percent. Furthermore, Moonshot reports that claim-level hallucinations on its internal stress test dropped from 23.4 percent to 8.3 percent, a notable improvement in reliability for production environments.

Specialized capabilities have also been expanded, particularly in technical domains. The model can interact directly with game engines such as Unity and Unreal Engine, handling both minor fixes and full project builds. In 3D modeling, it scores 92.7 percent on the BenchCAD benchmark when using tools, approaching the performance of top-tier rivals. Moonshot has also launched a "3D Make" feature on its platform, allowing users to preview and export complex architectural and mechanical designs.

A key strategic decision involves the release of model weights. Moonshot plans to make the weights available by November 4, 2026. If executed, Kimi K3.1 Pro would become the highest-scoring open-weights model on the market. This move would shrink the performance gap between open and proprietary models from 12 points to just three, potentially altering the cost-benefit analysis for organizations deciding between licensing and self-hosting.

However, the company has implemented deliberate constraints on offensive cybersecurity capabilities to manage risk for open-weights users. By removing specific cyber-related data from the training corpus, Moonshot reduced the model’s ability to solve vulnerability reproduction tasks from 90.5 percent to 81.0 percent on its internal benchmark. This is a calculated trade-off, aiming to prevent open models from reaching the absolute frontier of offensive cyber operations while still supporting defensive auditing tasks.

The stakes for the broader market are high. With Kimi K3.1 Pro available now through Kimi.ai and its API, US-based competitors face increased pressure to justify their price premiums. The combination of near-parity performance in agentic tasks and a significant price cut suggests that the era of paying a substantial premium for top-tier reasoning capability may be coming to an end.