AI News · Models ·
Moonshot AI launches Kim K3.1 Pro and plans to open its weights by November 4
Chinese AI company Moonshot AI released Kim K3.1 Pro, a post-trained upgrade of its 3 trillion parameter Kimi K3 model. It scored 65.7% on Terminal-Bench 4.0 and beat Claude Opus 5.5 and GPT-6 Astra on the OfficeQA Pro v2 financial analysis test, SiliconANGLE reported.
Key points
- Kim K3.1 Pro builds on Kimi K3, a 3 trillion parameter model released in July.
- It scored 65.7% on Terminal-Bench 4.0 and led rivals on OfficeQA Pro v2.
- Available via Moonshot's paid cloud now, with open weights planned by November 4.
What happened: Moonshot AI introduced Kim K3.1 Pro, a new flagship model that SiliconANGLE reported can perform some tasks better than Claude Opus 5.5 and GPT-6 Astra. It is a post-trained version of Kimi K3, which the Chinese company released in July.
The details: Kimi K3 has 896 individual neural networks with 3 trillion parameters and activates only 16 of them at a time to keep inference costs down. The company said post-training for the new model used significantly more processing power than its past projects. Kim K3.1 Pro scored 65.7% on the Terminal-Bench 4.0 coding benchmark, more than 7% ahead of Astra but slightly behind Opus 5.5. It scored 92.7% on BenchCAD, a 3D model coding test, within four percentage points of both rivals, and outperformed both on OfficeQA Pro v2, a test of complex financial analysis tasks.
Background: The model trails Astra and Opus 5.5 significantly on popular cybersecurity benchmarks. According to Moonshot AI, engineers deliberately removed cybersecurity-related information from the post-training data to address safety risks, and the company plans to add more advanced vulnerability analysis features to future models.
What to watch: Kim K3.1 Pro is available now through Moonshot AI's paid cloud services, and the company plans to open-source the weights by November 4. Pricing was not reported. Teams with data residency or vendor rules for Chinese providers will want to review those before testing the hosted version.
Our take
Teams weighing open-weight options get a new large candidate for coding and financial analysis, though it trails rivals on cybersecurity tasks.