Oct 28, 2026
Kimi K3.1 Pro beat GPT-6 Astra on a coding-agent test, Moonshot claims
Moonshot's Kimi K3.1 Pro is on its API and apps now, with full open weights promised by November 4. Its benchmark scores are Moonshot's own.
Sir Toastimir
Footage: Moonshot AI
What it is
Kimi K3.1 Pro is a new large model from Moonshot AI, released October 23. It can take in about one million tokens at once. It is available now through Moonshot's Kimi API, and Moonshot says the full weights will be released by November 4, 2026, so anyone can download and run it.
How it scores, per Moonshot
On Terminal-Bench 4.0, a test of AI agents doing real work in a computer terminal, Moonshot reports 65.7 percent. It lists GPT-6 Astra at 58.2 percent and Claude Opus 5.5 at 66.4 percent. These are Moonshot's numbers and have not been checked independently yet. Moonshot says the model matches or approaches the strongest closed models on many tests.
Hacking data removed
Because the weights will be public, Moonshot removed cyber-attack related training data. On its internal hacking test, xdev-bench-hard, it reports that a version without those cuts solved 90.5 percent of problems, versus 81.0 percent for the released model. Moonshot says its hacking ability remains behind closed frontier models that have been public for at least three months.
Visuals in the video
The demo clips and title image are from Moonshot's launch post. The bar charts were made by agentoast using Moonshot's reported numbers.
Links
Get your daily slice of AI
One short email with the best new AI tools, robots and AI & tech stock news. Free. Unsubscribe anytime.
We only use your email to send the newsletter.