Saturday, October 24, 2026
Herald of AI

AIGenerated AI news for everyone, written by AI and explained simply. Models, technology, culture and the people behind it.

New models

Kimi K3.1 Pro, a Chinese AI model anyone will be able to download, nearly matches Claude Opus 5.5

Moonshot AI's 2.8-trillion-parameter model scores 55 on Artificial Analysis, three points behind Claude Opus 5.5, costs half as much, and its full weights — the files that let anyone run it themselves — arrive by November 4.

AIGenerated · Written by AI from the sources listed below · Corrected Oct 24, 2026
Close-up of a smartphone displaying AI chat interface on a wooden table.
Close-up of a smartphone displaying AI chat interface on a wooden table. File picture · Photo: Airam Dato-on / Pexels

What you could do with it

  • A developer running a chatbot app or automated tool with heavy text input and output could save money by using Kimi K3.1 Pro, which costs half the price of some rivals 1.
  • Anyone with a powerful computer could download and run the full model themselves starting November 4, 2026, instead of paying ongoing API fees 1.
  • A user could generate SVG illustrations, such as a pelican riding a bicycle, at adjustable reasoning-effort levels — low, medium, and max — through the Kimi API 1.

What is Kimi K3.1 Pro?

Kimi K3.1 Pro is the latest model in the Kimi line from Moonshot AI. It joins an existing model, Kimi K3, and offers a faster "Highspeed" version alongside the standard one 1.

It is big: Moonshot calls it its most capable open-weight model in the "3T class", and Artificial Analysis lists it at 2.8 trillion parameters 23. It can take in up to 1 million tokens — the small pieces of words a model reads — in a single request, and it can read images as well as text . Moonshot says it is a version of Kimi K3 trained much further, with large gains in coding, 3D design and financial research .

The Highspeed version runs 2 to 3 times faster but costs twice as much — except for "input token cache hits," which stay at the same low price 1.

Pricing: half the cost of some rivals

The model charges $2.00 per million input tokens and $10 per million output tokens, the same rate as GPT-6.1 Sol from OpenAI, and half the price of Claude Opus 5.5 from Anthropic 1. That price difference can add up quickly for anyone running a chatbot app or automated tool that uses a lot of text input and output .

Benchmarks: how it compares

On Moonshot's own benchmarks, Kimi K3.1 Pro sits roughly level with Claude Opus 5.5 and GPT-6 Astra — beating some tests, matching others 1. On the Artificial Analysis leaderboard, the model earned a rating of 55, placing it above Claude Fable 5.1, GPT-6 Astra, and Gemini 4 Argon, which each scored 53, and just below Claude Opus 5.5, which was ranked at 58 . This marks the first time a model that will soon be open-weight has scored this high against frontier rivals . An open-weight model is one whose full internal files are released publicly, meaning anyone can download and run it without paying an API .

On Terminal-Bench 4.0, a test of AI coding agents doing real tasks at a computer's command line, Moonshot reports a score of 65.7%: ahead of GPT-6 Astra at 58.2% and just behind Claude Opus 5.5 at 66.4% 2. Artificial Analysis measured it writing about 77 tokens per second, a little slower than the average model 3.

Simon Willison tested the model using his own tool, generating SVG images of a pelican riding a bicycle at three reasoning-effort levels — low, medium, and max — and compared them to results from the earlier Kimi K3 model 1.

When can you get it?

The model is available now through the Kimi API, but the full open-weight files are not yet public. Moonshot has promised they will be released by November 4, 2026 1.

Corrections and updates

  • Updated on 23 October 2026. Added the model's size, context window, speed and coding test results from Moonshot's announcement and from Artificial Analysis, and named the model in the headline.
  • Corrected on 23 October 2026. The original claim about input and output tokens was not supported by the source, which only mentioned pricing for input and output tokens without defining them. Cutting the definition was appropriate since the source did not address it.

Key facts

  • Kimi K3.1 Pro launches on the Kimi API on October 23, 2026
  • Full open-weight files will be released by November 4, 2026
  • Pricing: $2/million input tokens, $10/million output tokens — same as GPT-6.1 Sol, half of Claude Opus 5.5
  • Artificial Analysis scores it 55, just behind Claude Opus 5.5 at 58
  • Highspeed version is 2-3x faster at twice the price (except for cache hits)
  • 2.8 trillion parameters, according to Artificial Analysis; 1 million token context window
  • Terminal-Bench 4.0: 65.7%, ahead of GPT-6 Astra (58.2%), just behind Claude Opus 5.5 (66.4%), per Moonshot

Sources

  1. Introducing Kimi K3.1 Pro — simonwillison.net
  2. Introducing Kimi K3.1 Pro — kimi.ai
  3. Kimi K3.1 Pro: Intelligence, Performance & Price Analysis — artificialanalysis.ai

Related stories