Summary
Kimi K3.1 Pro, the latest model from Moonshot AI’s Kimi team, is billed as the strongest open-weights model available, with open weights scheduled for release on November 4th. This video from David Ondrej walks through how it stacks up against frontier models from OpenAI and Anthropic, including Opus 5.5, GPT-6 Sol, Astra, and Fable 5.1.
The review covers published results on Terminal Bench 4.0 and SWE Marathon, along with the Artificial Analysis Intelligence Index, where Kimi K3.1 Pro ranks second behind Opus 5.5. Ondrej emphasizes cost per task rather than headline API pricing, noting that the model costs roughly half as much as Opus 5.5 and often finishes tasks more cheaply. He also highlights the large speed improvement over Kimi K3, with a faster high-speed mode available at a higher price.
Beyond the benchmarks, he runs his own tests inside an isolated Linux sandbox, including 3D modeling and CAD-style tasks using Blender and FreeCAD. The video also addresses controversies around the release, such as questions about whether the model relies on another model behind the scenes and claims of paid influencer promotion, and discusses where Kimi K3.1 makes sense compared with Opus and GPT-6 Sol, including options for self-hosting.
📺 Source: David Ondrej · Published October 28, 2026
🏷️ Format: Review
