New to this? Read it in simple words
- Anyone can now download Kimi K3.1 Pro, a very large open model from Moonshot AI.
- Its licence adds conditions for businesses with over $20 million in revenue.
- An outside tester scores it 46 overall, against 58 for the leader, Claude Opus 5.5.
- Tencent launched Hy4, but its final weights are only promised, not out yet.
- Weights
- The numbers a model learned in training; with them you can run it yourself.
- Open model
- A model whose weights anyone can download, often with a licence that sets conditions.
Open-weight models can be downloaded and run on your own computers, unlike models that work only through a company’s service. In October, Moonshot AI promised the weights of Kimi K3.1 Pro by November 4. Tencent is now promising the same for its new model.
Kimi’s weights are public
The weights are in a public Hugging Face repository that needs no approval. The 96 weight files total about 1.56 terabytes, far more than a normal laptop can run.
Moonshot’s model card lists 2.8 trillion parameters, with 104 billion active for each token (a word or part of a word). Artificial Analysis now lists it as an open-weights model.
Scores are Artificial Analysis’s Intelligence Index v5, from October 27.
The licence has limits
The licence is Moonshot’s own, not a standard open-source one. It lets anyone use, change and sell the model free of charge, with two main conditions.
A business that sells access to language models must sign a separate deal with Moonshot if it and its affiliates earned over $20 million in any 12 months.
Products with more than 100 million monthly users, or over $20 million in monthly revenue, must show “Kimi K3.1-Pro” prominently. Neither condition applies to internal use, or to use through Moonshot’s own products or certified partners.
Sources2
What the outside test says
Artificial Analysis changed its Intelligence Index on October 27. On version 5, Kimi K3.1 Pro scores 46, the highest for an open model. That replaces the 55 we reported in October, which came from the older index.
Claude Opus 5.5 leads at 58, and GPT-6 Astra has 57. Among models whose weights were already out on October 27, DeepSeek V4.1 Flash 1017 and GLM-5.3 led at 32.
One claim now has an outside result. Moonshot said it nearly matched Claude Opus 5.5 on Terminal-Bench 4.0, a coding test. Artificial Analysis ran that test and put Kimi K3.1 Pro first, at 67%. Claude Opus 5.5 leads FrontierCode 2.0, another coding test in the index.
Tencent’s Hy4
Tencent says Hy4 can read images and charts. It says the model can also, for the first time, work with command lines and graphical programs. It is in Tencent’s own apps and cloud, and on outside platforms such as OpenRouter.
Tencent “plans to release model weights and system cards in the coming days,” the company says. Until then, only the Hy4 preview from August is on Hugging Face, under an Apache 2.0 licence. Tencent has not stated a licence for the final model.
The test numbers are Tencent’s own. In a finance test run by its own teams, Hy4 scored 3.04, DeepSeek-V4.1-Flash-1017 scored 2.97 and Kimi K3.1 Pro scored 2.68. Artificial Analysis had no Hy4 page when we checked on November 4.
Tencent’s price is about $0.90 per million input tokens and $2.70 per million output tokens.
Hacking skills, and who checks them
Moonshot says it removed cyberattack material from training. Yet on its in-house test, the released model solved 81% of attack tasks, against 33.3% for K3.
Tencent says “offensive materials related to exploitation” were left out of Hy4’s training. Reuters lists reverse engineering and penetration testing among the filtered content, but Tencent’s release says those stayed in “for analysis purposes”.
Experts from Tencent’s own Xuanwu and Keen labs found “no significant concerns” about Hy4’s attack skills in the areas they checked, Tencent says. They are not outside checkers.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2
- 3Primary source · November 4, 2026Tencent Launches Hy4, Joining the Ranks of the World’s Leading Open-Source AI Models Tencent
- 4Primary source · October 23, 2026Kimi K3.1 Pro Tech Blog: Advancing Open Intelligence Moonshot AI (Kimi)
- 5
- 6Research · November 4, 2026Kimi K3.1 Pro: Intelligence, Performance & Price Analysis Artificial Analysis
- 7Research · November 4, 2026Tencent rolls out AI model trained with limited cybersecurity data amid hacking fears Reuters via The Star
- 8Research · November 4, 2026Tencent releases Hy4, a powerful frontier AI model introducing visual understanding and computer use SiliconANGLE
- 9
- 10
The outside numbers for Kimi come from Artificial Analysis, and Tencent’s numbers are all its own. Reuters and Tencent describe Hy4’s filtered training data differently, so we quote Tencent’s text.
Kimi K3.1 Pro’s weights are out, and Tencent launches Hy4. Silicon AI News, November 4, 2026. https://siliconainews.com/stories/kimi-k3-1-pro-weights-tencent-hy4
Kimi K3.1 Pro’s weights are public as the top open model and Tencent launched Hy4, though licence limits and stronger hacking skills temper this. How we rate the mood
The mood shows how the news frames the event for the tech and AI world: as an opportunity (6 and above) or as a loss (below 5). It is not our verdict on who is right.