Summary
Moonshot AI has released Kimi K3.1 Pro, an upgrade to its July Kimi K3 model, and early benchmark results put it in contention with Claude Opus 5.5 and GPT-6 Astra. This video from TheAIGRID walks through what the model is, what it can do, and whether the numbers hold up.
The model has roughly 2.8 trillion total parameters, using a mixture-of-experts design that activates about 104 billion per token, and supports a 1 million token context window with image understanding. Moonshot reports big jumps over K3, including Terminal Bench rising from 13% to 65.7% (against 66.4 for Opus 5.5) and a climb from 17th to 8th on Arena’s web development leaderboard. It scored 92.7% on Bench CAD with tools. The video also covers Moonshot’s plan to release open weights, the Kimi Work and Kimi Code products, and a faster High Speed variant.
Pricing is a major theme: $2 per million input tokens and $10 per million output, roughly half the cost of Opus 5.5 and about a fifth of GPT-6 Astra. A Moonshot demo of a 3D rocket launch scene showed Kimi finishing in about 6 minutes for $1.95 versus about 24 minutes and $10.35 for Opus. The host also addresses skepticism that the model is over-optimized for benchmarks, and argues that good-enough performance at a fraction of the price may matter more to many users and businesses.
📺 Source: TheAIGRID · Published October 23, 2026
🏷️ Format: News Analysis
