Guotai Haitong has released a research note stating that Moonshot AI launched Kimi K3.1 Pro in the early hours of October 24, ranking fourth globally on an independent intelligence index and just 3 points behind the closed-source frontier; post-training compute has been significantly increased, with the model topping multiple leaderboards in long-horizon coding and Agent benchmarks; API pricing has been cut by 33% compared to K3, further clarifying the commercialization path for open-source models, and the brokerage is bullish on how a thriving open-source model ecosystem will accelerate the deployment of AI applications.
Kimi K3.1 Pro Launches, Open Weights Approach the Closed-Source Frontier
At 1:16 a.m. on October 24, Moonshot AI released Kimi K3.1 Pro, the first model in the K3.1 series, through its official WeChat account. As a post-training version of K3, it is the most capable 3-trillion-parameter model to date, supports a 1-million-token context, and remains the world's largest open-source model by parameter count.
On the Artificial Analysis intelligence index, K3.1 Pro ranks fourth globally with a score of 55, an 11-point improvement over K3 and only 3 points behind the top-ranked Claude Opus 5.5. Once its weights are opened, it will become the highest-scoring open-weight model globally, narrowing the gap between open-source and closed-source frontiers from 12 points to 3 points. Full model weights are scheduled for release before November 4.
Post-Training Compute Significantly Increased, Topping Multiple Long-Horizon Coding and Agent Benchmarks
According to official information, to support large-scale joint reinforcement learning across hundreds of downstream domains, K3.1 Pro underwent full-stack optimization of its reinforcement learning framework, long-context reasoning, training efficiency, sandbox infrastructure, and deployment scheduling.
It achieved 65.7% on Terminal-Bench 4.0, surpassing GPT-6 Astra's 58.2% and trailing only Claude Opus 5.5's 66.4%. It ranked first on SWE-Marathon v1.1 with 51.7%, ahead of Claude Opus 5.5's 46.3%. It topped the officially compared models on Agents' Last Exam with 43.2 points. Across coding, Agent, and knowledge benchmarks, it reached or approached the level of the strongest closed-source models.
API Pricing Cut 33% Versus K3, Enabling End-to-End Deployment in 3D Creation and Financial Workflows
K3.1 Pro API pricing is set at US$2 per million tokens for input and US$10 per million tokens for output, with all three rate tiers reduced by 33% compared to K3. A Highspeed version was also launched simultaneously, offering roughly 2 to 3 times faster output at equivalent intelligence for twice the standard version's price.
The model can directly write and run scripts for professional modeling tools such as Blender and CAD, and the 3D Make feature has been launched simultaneously on kimi.com and the Kimi App. In the financial sector, it delivers end-to-end capabilities from retrieval and verification to financial modeling and profitability analysis, achieving the current best score of 64.4 on the OfficeQA Pro v2 benchmark.
Open-weight models are approaching closed-source models on both capability and cost fronts, further opening up the commercialization space for the open-source ecosystem. Risks: technological development falling short of expectations and intensifying market competition.
Comments