← Latest AI News

October 24, 2026 · Daily brief

Frontier AI at half the price—starting today

Sovereignty angle
A Chinese lab just matched the top Western models at half the price. That is the move: test it this week on work you can check, keep two vendors live, and never send data you cannot afford to lose.

Kimi K3.1 Pro from China's Moonshot AI costs $2 in, $10 out per million tokens and scores 55 on independent tests—three points behind Claude Opus 5.5, half the price of GPT-6.

What changed

Moonshot AI released Kimi K3.1 Pro on October 23. It costs $2 per million input tokens and $10 per million output. That is half what OpenAI charges for GPT-6 and matches GPT-6.1 Sol exactly. Artificial Analysis scored it 55 out of 100. Only three models scored higher: Claude Opus 5.5 at 58, Claude Sonnet 5.5 at 56, and Meta Muse Wonder Full at 56. Moonshot promises full open weights by November 4.

What you can do with it this week

Run the same tasks you give to ChatGPT or Claude through Kimi instead. The API works with any tool that speaks OpenAI format. Try it on your longest recurring job first: the client proposal you rewrite every week, the monthly report you pull from three spreadsheets, the follow-up email sequence your team sends after every call. A task that burned 10 million tokens on GPT-6 at $60 now costs $30 on Kimi. Set a two-week test. Track what it gets right and where you still edit. If quality holds, cut your API bill in half.

What to keep in your own hands

Do not move your entire operation to one vendor because the price dropped. Keep at least two providers live. Test Kimi with data you can verify: if it hallucinates a number in a financial summary, you need to catch that before the client does. The open weights arrive November 4; until then, your data still crosses Moonshot servers. Never paste in customer lists, contract terms, or anything you would not hand a stranger on the street.