Several large AI labs outside the United States publish models as open weights that anyone can download. DeepSeek, Alibaba’s Qwen team, Moonshot AI, Z.ai, MiniMax, Xiaomi, Tencent and Mistral publish big models this way; ByteDance publishes some models but keeps its Seedance video model closed. What “open” lets you do differs from one licence to the next, and that is the part worth reading before you build on any of them.
- The plain answer: you can download and run many of these models yourself, for free, but each one comes with its own rules. Read the licence file before a business relies on one.
- All ten labs here publish some weights. Each does so on its own Hugging Face page; ByteDance keeps Seedance closed, and StepFun’s newest model, Step 5, had no official download as of 1 Oct 2026. The list is a snapshot; the dated rankings are in their own section.
- “Open weights” is not one licence. Some are MIT or Apache 2.0. Others add conditions once you are large, cap company revenue, or allow only non-commercial use by default.
- One video licence excludes four territories by default. MiniMax’s H3 licence does not cover the EU, the UK, South Korea or the United States; people there must apply to MiniMax for a licence.
- The hosted app and the downloaded file are different things. A lab’s privacy policy governs its own service, not weights you run on your own hardware. Kimi K3 can also be used through Amazon Bedrock.
- Download only from the lab’s own page. Lookalike repositories with a famous model’s name exist, including copies claiming to be Step 5.
Weights are the model’s files: the numbers it learned in training. Open weights means the lab publishes those files, so anyone can download the model and run it on their own computer or server. Hugging Face is the website where most labs publish them. An API is the other route: you send your text to the lab’s own servers and get the answer back, so the lab runs the model and sees what you send.
Why open weights matter here
This site separates frontier labs whose models you rent from an open-weights layer whose models you can download, run and keep, trailing the frontier by months, not years. That layer is why AI is not simply owned by a few companies; see who builds AI and open or closed. Open weights are not open source: you can run and adapt the model, but not rebuild it.
A September 2026 example shows the difference, and how little warning an API gives. On 10 September DeepSeek announced that “Starting at 04:00 UTC on Sept 14, 2026, all deepseek-v4-pro requests will route to V4.1-Flash at V4.1-Flash rates.”1 Before that date arrived, DeepSeek reversed the plan. Its change log says: “In response to user demand, we have decided to continue providing API services for DeepSeek V4 Pro after September 14, 2026, with the billing method remaining unchanged.” Its pricing page still lists deepseek-v4-pro, served by DeepSeek-V4-Pro-0813, at prices of its own.2 So anyone calling that API name had days’ notice of a model swap they had not asked for, and then had to catch a second notice saying it was off. Anyone who downloaded the V4-Pro weights did not need to read either one.Reasoning, September 2026 — follows from the two notices above and from the V4-Pro weights remaining on DeepSeek’s Hugging Face page, read at source 22 Sep 2026.First-hand: until 22 Sep 2026 this paragraph said “Anyone calling that API name now gets a different model.” That was wrong. DeepSeek had withdrawn the switch in its change log before it was due; this page took the plan from the 10 Sep news post, which still carries the original sentence and not the reversal.
An API can be changed under you with a few days’ notice, and changed back just as fast. A file on your own disk cannot. That is the practical reason these labs matter to anyone who builds on AI.
The labs, one at a time
DeepSeek (China). Its privacy policy names the operator as “Hangzhou DeepSeek Artificial Intelligence Co., Ltd., with its registered address in China”.3 Its newest release, DeepSeek-V4.1-Flash, is dated 10 September 2026, and its model card’s licence section reads: “This repository and the model weights are licensed under the” MIT License.4 It was still the newest entry in DeepSeek’s change log on 1 October 2026.5
Qwen (Alibaba). Alibaba released the weights of its flagship on Hugging Face and on ModelScope, which it describes as “an Alibaba-initiated open-source community”. Its release log dates Qwen3.8-2.4T-A95B to 12 August 2026 and the smaller Qwen3.8-27B to 14 August, and says: “For the first time, Qwen3.8 brings a Qwen-Max-class model to open release.”6 The two carry different licences, and a later model, Qwen3.8-Flash-Next, carries a third; see the next section. The next generation is not out: at its Apsara conference on 22 September, Alibaba said “Next-generation Qwen 4 is in training”.7
Moonshot AI (Kimi). Its About page gives a Beijing address and says the company was founded in spring 2023. It lists Kimi K3 on 16 July 2026, “a 2.8T-parameter model” released “under the Kimi K3 License”.8 Since 18 September 2026 Kimi K3 has also been on Amazon Bedrock, Amazon’s cloud service for AI models. Amazon runs that copy on its own servers, and offers a US-only option: its us.moonshotai.kimi-k3 profile “routes requests only among US-geography Regions to respect US data residency”.9
Z.ai (GLM models). Its GitHub page links each GLM download on ModelScope under the organisation name ZhipuAI, and its privacy policy says the international service is run from Singapore. It publishes GLM-5.3 and a smaller GLM-5.3-Flash, “With 320B total parameters and just 18B active parameters”. (Parameters are the model’s learned numbers; “active” ones are the share it uses for each word, which largely sets how much computing each answer takes.)10
MiniMax. It calls itself “a global AI foundation model company. Founded in early 2022”. It publishes MiniMax-M3, a multimodal model, and MiniMax H3, a video model that “can generate video with native stereo audio at resolutions up to 2K and durations of up to 15 seconds”. The H3 licence gives its release date as 2 August 2026 and is “governed by the laws of the Hong Kong Special Administrative Region”.11 MiniMax’s own release notes date MiniMax M3 to 1 June 2026 and H3 to 31 July 2026.12
ByteDance Seed. Seedance 2.5, its video model, is “built for 30-second storytelling with precise reference control and powerful editing capabilities”; its page offers “Get API” and “Try now”, not a download. The ByteDance-Seed organisation on Hugging Face listed 59 models on 18 September 2026, none of them Seedance. It does publish Seed-OSS-36B, a language model, under Apache 2.0.13 Prompting it is covered on the Seedance cheatsheet.
Xiaomi (MiMo models). Added 1 October 2026. The phone maker released MiMo-V2.6-Pro and MiMo-V2.6-Flash on 22 September 2026 and says: “We have fully open-sourced the weights and technical report of the MiMo-V2.6-Pro and Flash models”. Its announcement names no licence; the V2.6 repositories on its Hugging Face page are tagged MIT.14 Xiaomi also calls it “the most powerful open-source model available”; that is the company’s claim, and the independent rankings are below.
Tencent (Hy models, formerly Hunyuan). Added 1 October 2026. Tencent published a preview of Hy4, a 770-billion-parameter model, on Hugging Face; the repository dates from 27 August 2026. Its card says: “Hy4 preview is released under the Apache License 2.0.”15
StepFun. Added 1 October 2026. StepFun’s newest model, Step 5 Preview, is on its API; its own documentation says it “natively supports text, image, and video input” and says nothing about weights. MarkTechPost reported on 20 September that open weights were scheduled for 15 October 2026. On 1 October the official stepfun-ai page on Hugging Face still had no Step 5 files; its newest was Step-3.7-Flash, from May 2026.16 Some other Hugging Face accounts already offer files named Step-5-Preview; see where to find them safely.
Mistral AI (France). Its privacy policy says “Mistral AI is a French company incorporated in Paris”. Mistral Medium 3.5, announced on 22 May 2026, was “Released as open weights, under a modified MIT license.” Mistral Small 4 is under Apache 2.0.17
| Lab | Newest open weights | Licence on the weights | Official download page |
|---|---|---|---|
| DeepSeek | DeepSeek-V4.1-Flash | MIT | huggingface.co/deepseek-ai |
| Qwen (Alibaba) | Qwen3.8-27B; Qwen3.8-2.4T-A95B; Qwen3.8-Flash-Next | Apache 2.0; Qwen3.8-Max License; Qwen Community License 1.0 | huggingface.co/Qwen, ModelScope |
| Moonshot AI | Kimi K3 | Kimi K3 License | huggingface.co/moonshotai |
| Z.ai | GLM-5.3; GLM-5.3-Flash | GLM-5.3 License; MIT | huggingface.co/zai-org |
| MiniMax | MiniMax-M3; MiniMax H3 (video) | MiniMax Community License; H3 Community License | huggingface.co/MiniMaxAI |
| ByteDance Seed | Seedance 2.5: no weights. Seed-OSS-36B | Seed-OSS: Apache 2.0 | huggingface.co/ByteDance-Seed |
| Mistral AI | Mistral Medium 3.5; Mistral Small 4 | Modified MIT; Apache 2.0 | huggingface.co/mistralai |
| Xiaomi | MiMo-V2.6-Pro; MiMo-V2.6-Flash | MIT (Hugging Face tag) | huggingface.co/XiaomiMiMo |
| Tencent | Hy4 preview | Apache 2.0 | huggingface.co/tencent |
| StepFun | Step-3.7-Flash. Step 5 Preview: no official weights as of 1 Oct 2026 | Apache 2.0 (Hugging Face tag) | huggingface.co/stepfun-ai |
Licence names from each model’s own card and LICENSE file on Hugging Face, and the per-organisation listings on Hugging Face, all read at source 18 Sep 2026. The Xiaomi, Tencent and StepFun rows were added 1 Oct 2026 from the sources in their paragraphs above (Tencent read 29 Sep, Xiaomi and StepFun 1 Oct 2026). Step-3.7-Flash licence: Hugging Face, API record for stepfun-ai/Step-3.7-Flash, read 1 Oct 2026: “license:apache-2.0”.
Ten labs, and one of them keeps its video model closed. Check each model separately; the same lab can release under two different licences in the same month.
“Open weights” is not one licence
MIT and Apache 2.0 are standard permissive licences that allow commercial use, with conditions such as keeping the licence notice. The other labs write their own terms, and those differ in ways that matter to a business.18
Conditions once you are large. The Qwen3.8-Max licence, covering the 2.4-trillion-parameter model, wants the model name shown in products with “more than 100,000,000 monthly active users or US$ 20,000,000 (or equivalent in other currencies) monthly revenue”. A model API or work-assistant business whose “aggregate revenue of the licensee and its affiliates exceeds US$50,000,000” in twelve months must “obtain a separate license from Qwen before Using the Software or its derivative works for any commercial purpose”. The Kimi K3 licence sets its API-business threshold at 20 million US dollars.19First-hand: until 22 Sep 2026 this paragraph quoted the display clause only as far as “more than 100,000,000 monthly active users”, leaving out its monthly-revenue alternative. Full clause re-read in the Qwen3.8-Max License at source 22 Sep 2026. Alibaba’s announcement stresses Apache 2.0, which is true of the 27B model, not the larger one; read the file next to the weights, not the headline.Reasoning, September 2026 — compares the Alibaba Cloud post (“Released under the Apache 2.0 license”, said of Qwen3.8-27B) with the two model cards, all read at source 18 Sep 2026.
A separate deal for any API business. Qwen3.8-Flash-Next, on Hugging Face since 24 August 2026, carries a third Qwen licence, the Qwen Community License 1.0. Its display clause is the same as the Qwen3.8-Max one. Its second clause drops the revenue threshold: if the licensee “conducts a Model as a Service or AI Work Assistant business, the licensee shall obtain a separate license from Qwen before Using the Software or its derivative works for any commercial purpose.” Internal use that makes nothing available to third parties is exempt.20
A security review. The GLM-5.3 licence adds one condition beyond keeping the notice: a company running a model API business with more than 10 billion US dollars of revenue over twelve months must pass Z.ai’s security review before commercial use. “The scope and method of the security review shall be reasonably determined by Z.AI.” GLM-5.3-Flash, by contrast, is MIT.21
A revenue cap. Mistral Medium 3.5’s modified MIT licence gives no rights at all to larger companies: “You are not authorized to exercise any rights under this license if the global consolidated monthly revenue of your company (or that of your employer) exceeds $20 million (or its equivalent in another currency) for the preceding month.”22
Non-commercial by default, and limited by territory. The MiniMax-M3 licence grants rights “to deal in the Software for non-commercial purposes”; commercial use requires crediting MiniMax and a notice to it, or its written authorisation above 20 million US dollars of yearly revenue. The H3 video licence goes further: its “Excluded Territories” mean “the European Union, the United Kingdom, the Republic of Korea and the United States of America”.23
Research only. Update, 22 September 2026: Qwen’s image model Qwen-Image-2.1 is under none of the three Qwen licences above. Its Qwen Research License Agreement, marked “Release Date: September 20, 2026”, grants rights “FOR NON-COMMERCIAL PURPOSES ONLY” and defines “Non-Commercial” as “for research or evaluation purposes only”. Commercial use needs “a separate commercial license” from the licensor, Hangzhou Tongyi Laboratory Technology Co., Ltd. The same lab now publishes under Apache 2.0 and under research-only terms.24
Before building on an open model, open its LICENSE file. “Open weights” tells you the file is downloadable; only the licence tells you whether your company may use it.
Where to find them safely
Each lab above publishes its weights under its own Hugging Face organisation, listed in the table; Qwen, Z.ai and MiniMax also link copies on ModelScope. Start from the lab’s page and follow its link. A Hugging Face search for “seedance” on 18 September 2026 returned 17 model repositories, none owned by ByteDance-Seed.25 A famous name on a repository does not make it the lab’s release. For running a model yourself, see local or cloud.
A live example, 1 October 2026: a Hugging Face account called TypeSafeAI offers “Step-5-Preview-BF16” and says “BF16 checkpoint available now”. Its card uses StepFun’s name and links, but it is not on StepFun’s own stepfun-ai page, which had no Step 5 files that day.26 This site has not checked what those files contain. The point is where they come from: not from StepFun. More on staying safe is in is DeepSeek, Kimi or Qwen safe to use?
Check the organisation name before the model name. If the maker has not published weights, a download claiming to be that model is not from the maker.
What is genuinely uncertain
The hosted apps and APIs are separate from the weights, and their policies differ. DeepSeek’s says: “The Personal Data we collect from you may be stored on a server located outside of the country where you live.” It names the People’s Republic of China as where it stores that data, and lists “the right to opt-out of using your Personal Data for training our models or optimizing our technologies.”3 Z.ai’s international service says data “is generally processed in Singapore”. Kimi’s names NOVASCENT PRIVATE LIMITED as “the provider and controller” and says user content is processed to improve the services, “including training and optimizing our artificial intelligence” models.27 These policies cover each lab’s own service, not weights you run on your own hardware.Reasoning, September 2026 — a privacy policy covers the lab’s own service; see what happens to what you type.
The one territorial restriction found in these primary sources is MiniMax’s, and the licence gives no reason for it. It says: “We will continuously evaluate the applicable laws, regulations and compliance requirements for the Excluded Territories.” The same clause goes on to offer a way in: anyone in those territories “interested in deploying our models” is “welcome to contact us about obtaining a license, which will be granted based on robust controls and guardrails”. The H3 model card links an “Application form(only for USA/EU/UK/South Korea)”.28First-hand: until 1 Oct 2026 this paragraph said the licence “says only” that MiniMax will keep evaluating the laws. That was incomplete: the same clause invites people in the excluded territories to apply for a licence, and the model card links an application form. This page does not guess at the reason.
Read two documents: the licence for the weights, and the privacy policy for the app. They answer different questions, and one does not stand in for the other.
Which ranks highest, by independent tests
The plain answer: Xiaomi, Z.ai and Moonshot sit at the top of the open-weights tables, with little between them. These are scores from outside groups, not from the labs. They move every few weeks, so each carries the date it was read.
Artificial Analysis combines many tests into one “Intelligence Index” (version 4.3.2). On its open-weights page, read on 1 October 2026: MiMo-V2.6-Pro 46, GLM-5.3 (max) 45, Kimi K3 (max) 44, GLM 5.3 Flash 42, DeepSeek V4.1 Flash 39, Qwen3.8 27B 34, MiniMax-M3 29. Scores from other index versions cannot be compared with these.29
Arena ranks models by people’s votes in blind side-by-side tests. On its text board dated 25 September 2026, the best-placed model from these labs was kimi-k3-max at 16th (1488), behind models from Anthropic, Google and others; the top model was claude-opus-5.5-high (1509). Next came mimo-v2.6-pro 23rd and glm-5.3-max 24th (both 1480), qwen3.8-max 25th (1479) and deepseek-v4.1-flash-max 29th (1477). Gaps this small sit inside the board’s own margins of error.30 A fuller comparison, lab by lab, is on Chinese AI models, explained.
The best downloadable models are close together and close behind the best closed ones. Re-check the boards before choosing; the order changes monthly.
What this page could not verify
- The labs’ own benchmark claims. Each card compares itself with rivals; none of those scores was checked here. The independent rankings above are dated snapshots.
- Training data. This page did not assess what any of these models was trained on.
- Where MiniMax and ByteDance Seed are based. The pages read here give no headquarters address for either.
- The release date for Mistral Small 4, and the exact day the GLM weights went up. Z.ai’s developer release notes list “2026-08-18 GLM-5.3” and “2026-08-26 GLM-5.3-Flash”; those are API release dates. Hugging Face shows both GLM repositories were created on 25 August 2026. That is the day the repository was made, which need not be the day the files went public. MiniMax-M3’s date is now known: 1 June 2026, from MiniMax’s release notes (see above). For Mistral Small 4, no date was found on a page that could be read.31
- Why MiniMax excludes four territories. The licence does not say.
- When StepFun will publish Step 5 weights. The 15 October date comes from a news report, not from StepFun’s own pages.
- Whether this list is complete. Other labs outside the US publish weights; this page covers ten that could be checked at source.
Related on this site: Can You Run Chinese AI Video Models on Your Own Computer?
- 1DeepSeek, DeepSeek-V4.1-Flash: Smarter, Faster, More Efficient, 10 Sep 2026, read at source 22 Sep 2026“Starting at 04:00 UTC on Sept 14, 2026, all deepseek-v4-pro requests will route to V4.1-Flash at V4.1-Flash rates.” The news post still carries this sentence. ↑
- 2DeepSeek, Change Log, entry dated 2026-09-10, and Models & Pricing, both read at source 22 Sep 2026. Internet Archive copies of the change log lack the reversal on 11 Sep 2026 and carry it on 13 Sep 2026.↑
- 3DeepSeek, Privacy Policy, last updated 10 Feb 2026, read at source 18 Sep 2026.↑a↑b
- 4DeepSeek, API news: “DeepSeek-V4.1-Flash Release 2026/09/10”; DeepSeek-V4.1-Flash model card. Both read at source 18 Sep 2026.↑
- 5DeepSeek, Change Log, newest entry dated 2026-09-10, read at source 1 Oct 2026“The previous-generation models V4 Flash and V4 Flash Vision Exp have been retired”. ↑
- 6Alibaba Cloud, Alibaba Unveils Qwen3.8-27B and Releases Weights of Qwen3.8 Flagship Model; Qwen, Qwen3.8 on GitHub: “2026-08-12: Qwen3.8-2.4T-A95B is now available on” and “2026-08-14: Qwen3.8-27B is now available on”. Both read at source 18 Sep 2026.↑
- 7Alibaba Cloud, Apsara 2026 blog post, dated 25 Sep 2026, read at source 29 Sep 2026“Next-generation Qwen 4 is in training”. ↑
- 8Moonshot AI, About: “Address 76 Zhichun Road, Haidian District, Beijing JD Technology Building 13F”; home page: “2026-07-16 Kimi K3”; Kimi-K3 model card. All read at source 18 Sep 2026.↑
- 9AWS, What’s New announcement for Kimi K3 on Amazon Bedrock, posted 18 Sep 2026, read at source 1 Oct 2026“Kimi K3 from Moonshot AI is generally available on Amazon Bedrock”; AWS, Kimi K3 model card in the Bedrock user guide, read at source 1 Oct 2026. ↑
- 10Z.ai, GLM-5 on GitHub, linking “https://modelscope.cn/models/ZhipuAI/GLM-5.3”; GLM-5.3-Flash model card; Privacy Policy: “We generally provide the Services from Singapore”. All read at source 18 Sep 2026.↑
- 11MiniMax, About; MiniMax-H3 model card; MiniMax H3 Community License Agreement: “MiniMax H3 release date/License date: August 2, 2026.” All read at source 18 Sep 2026.↑
- 12MiniMax, Models release notes, read at source 1 Oct 2026“Jun. 1, 2026” MiniMax M3; “Jul. 31, 2026” MiniMax H3, “A new-generation open general-purpose multimodal video model.” ↑
- 13ByteDance Seed, Seedance 2.5, read in a browser 18 Sep 2026; Hugging Face, ByteDance-Seed model listing, read at source 18 Sep 2026; Seed-OSS-36B-Instruct model card: “We release this series of models to the open-source community under the Apache-2.0 license.”↑
- 14Xiaomi, MiMo-V2.6 release note, 22 Sep 2026, read at source 1 Oct 2026; Hugging Face, API listing for XiaomiMiMo, read 1 Oct 2026MiMo-V2.6-Pro-RL and MiMo-V2.6-Flash-RL, created 2026-09-21, “license:mit”. ↑
- 15Tencent, Hy4-preview model card and LICENSE, read at source 29 Sep 2026.↑
- 16StepFun, Step 5 Preview model guide, read at source 29 Sep 2026; MarkTechPost, report on Step 5 Preview, 20 Sep 2026, read at source 29 Sep 2026; Hugging Face, API listing for stepfun-ai, read 1 Oct 2026newest “stepfun-ai/Step-3.7-Flash”, created 2026-05-23. ↑
- 17Mistral AI, Privacy Policy; Remote agents in Vibe. Powered by Mistral Medium 3.5, dated May 22, 2026; Mistral Small 4 model card: “Open-source license for both commercial and non-commercial use”. All read at source 18 Sep 2026.↑
- 18Mistral AI, Mistral Small 4 model card, read at source 18 Sep 2026, describing Apache 2.0 as “Open-source license for both commercial and non-commercial use”.↑
- 19Qwen, Qwen3.8-Max License; Moonshot AI, Kimi K3 License: “the aggregate revenue of the Licensee and its affiliates exceeds 20 million US dollars”. Both read at source 18 Sep 2026.↑
- 20Qwen, Qwen Community License 1.0 for Qwen3.8-Flash-Next, read at source 22 Sep 2026; date from the Hugging Face API listing for Qwen, read 22 Sep 2026“2026-08-24T08:24:59.000Z Qwen/Qwen3.8-Flash-Next”. ↑
- 21Z.ai, GLM-5.3 License: “exceeds 10 billion US dollars (or the equivalent in other currencies) in total over any consecutive 12 months”; GLM-5.3-Flash model card: “license: mit”. Both read at source 18 Sep 2026.↑
- 22Mistral AI, Modified MIT License, read at source 18 Sep 2026“exceeds $20 million (or its equivalent in another currency) for the preceding month”. ↑
- 23MiniMax, MiniMax Community License: “if such products and services generate more than 20 million US dollars (or equivalent in other currencies) in yearly revenue”; MiniMax H3 Community License Agreement. Both read at source 18 Sep 2026.↑
- 24Qwen, Qwen Research License Agreement and Qwen-Image-2.1 model card, both read at source 22 Sep 2026.↑
- 25Hugging Face, model search for “seedance”, read through the Hugging Face API 18 Sep 2026.↑
- 26Hugging Face, TypeSafeAI/Step-5-Preview-BF16, read at source 1 Oct 2026“Open Weights: BF16 checkpoint available now”; API listing for stepfun-ai, read 1 Oct 2026. ↑
- 27Z.ai, Privacy Policy; Kimi, Privacy Policy, last updated 13 Aug 2026. Both read at source 18 Sep 2026.↑
- 28MiniMax, MiniMax H3 Community License Agreement, read at source 18 Sep 2026 and again 1 Oct 2026; MiniMax-H3 model card, read at source 29 Sep 2026.↑
- 29Artificial Analysis, open-weights models page, Intelligence Index v4.3.2, no date on the page, read at source 1 Oct 2026“MiMo-V2.6-Pro and GLM-5.3 (max) are the highest intelligence open source models, followed by Kimi K3 (max) & GLM-5.3-Flash.” ↑
- 30Arena, Text leaderboard, last updated 25 Sep 2026, read at source 1 Oct 2026“16. kimi-k3-max, 1488”; kimi-k3-max shown as 1488±5 on the open-source view, read 1 Oct 2026. ↑
- 31Z.ai, New Released release notes, read at source 22 Sep 2026; Hugging Face, API listing for zai-org, read at source 29 Sep 2026GLM-5.3 and GLM-5.3-Flash created 2026-08-25. ↑
DeepSeek, V4.1-Flash release notice, Change Log, Models & Pricing, model card and Privacy Policy · Alibaba Cloud, Qwen3.8 announcement; Qwen, GitHub, Qwen3.8-27B, Qwen3.8-Max License, Qwen Community License 1.0 and Qwen Research License Agreement · Moonshot AI, About, Kimi-K3 card, Kimi K3 License and Kimi Privacy Policy · Z.ai, GLM-5 GitHub, GLM-5.3 License, GLM-5.3-Flash card, release notes and Privacy Policy · MiniMax, About, MiniMax Community License and MiniMax H3 Community License Agreement · ByteDance Seed, Seedance 2.5 and Seed-OSS-36B-Instruct · Mistral AI, Privacy Policy, Mistral Medium 3.5 announcement, Modified MIT License and Mistral Small 4 card · Hugging Face organisation listings and search. Read at source on 18 September 2026; the DeepSeek change log and pricing page, the three Qwen licences and Z.ai’s release notes read on 22 September 2026.
Added 1 October 2026: MiniMax, Models release notes and MiniMax-H3 card · Alibaba Cloud, Apsara 2026 post · AWS, Kimi K3 on Amazon Bedrock announcement and Bedrock model card · Xiaomi, MiMo-V2.6 release note · Tencent, Hy4-preview card · StepFun, Step 5 Preview guide; MarkTechPost, Step 5 Preview report · Artificial Analysis, open-weights models · Arena, Text leaderboard · Hugging Face listings for XiaomiMiMo, stepfun-ai and zai-org, and the TypeSafeAI Step-5 copy. Read at source 29 Sep or 1 Oct 2026, as marked beside each claim.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK