Moonshot Unleashes Kimi 3.1 Pro: A New Open-Weight AI Challenger Rivalling Frontier Models
Moonshot has introduced Kimi 3.1 Pro, an open-weight model and the inaugural offering in its K3.1 series. This new model features 3 billion parameters and boasts an impressive 1 million token context window, marking it as the company’s largest release to date. Concurrently, Moonshot launched a “High Speed” inference option, promising 2-3 times faster processing at twice the price. Early benchmarks indicate Kimi 3.1 Pro’s performance is closely aligned with, and in some metrics, surpasses, models like Opus 5, while consistently competing with GPT-6 Astra and Fable. Third-party evaluations further bolster these claims; cybersecurity benchmarks place Kimi 3.1 Pro at number two, ahead of Deepsic and Grok 4.7, and Artificial Analysis ranks it fourth overall, outperforming Fable 5.1, GPT-6 Astra, and Gemini 4 Argon. The model’s API pricing is set at $2 per million input tokens and $10 per million output tokens, representing a 35% cost reduction compared to its K3 predecessor.
Kimi 3.1 Pro demonstrates robust capabilities in 3D content generation, exhibiting strong performance with popular engines such as Unity, Unreal Engine, Godot, and 3GS, as showcased by generated game demos and interactive landing pages. One developer’s experience highlighted Kimi 3.1’s practical application in creating a full 3D, browser-based game, “Halloween.Dev,” within a week, leveraging the model for core development and optimizing 3D assets through iterative feedback. Despite its compelling performance and competitive pricing, Moonshot has acknowledged challenges related to rate limitations and unnecessary quota consumption, offering subscribers a 25% monthly quota reset. This points to potential computational infrastructure constraints in meeting high demand, a common hurdle for many emerging AI firms. The open weights for Kimi 3.1 Pro are anticipated for release in November.