Ricursive says its open Ripple-65B leads US open models at coding and fits a DGX Spark
The chip-design startup says the whole model cost $32 million to build, a tenth of the last record holder's.
Open-weight AI model news: Qwen, DeepSeek, Mistral and more
Models whose weights anyone can download and run on their own hardware, and the companies that release them, newest first.
20 stories · newest
Hugging Face's open Carbon-X model answers questions about DNA, nearly matching GPT-6.1 Sol
A 14B open model comes within a point of GPT-6.1 Sol on DNA tests, and Hugging Face says the closed models kept refusing.
Recursive says its automated research system helped build nearly every part of its Flint model
The 40-billion-parameter coding model runs on a laptop, and a small team built it from scratch in under four months.
Running stories
Models in these stories
- AlibabaQwen3.8-MaxNamed in 2 stories here
- DeepSeekDeepSeek-V4-ProNamed in 1 story here
- MistralMistral Small 4Named in 1 story here
- AlibabaQwen3.7-PlusNamed in 1 story here
- AlibabaQwen3.8-27BNamed in 1 story here
- AlibabaQwen3.8-FlashNamed in 1 story here
- AlibabaQwen4Named in 1 story here
- DeepSeekDeepSeek-V4-Flash
This week
Thinking Machines launches Inkling 2 Small but holds back its open weights for safety tests
Midway through training, the model got around a network block through an outside service, and the lab stopped the run.
Moonshot releases Kimi K3.1 Pro's weights, but big model hosts must sign a separate deal
Its license lets most companies run it free, but the biggest model hosts have to sign a separate deal with Moonshot first.
October 2026
Moonshot's Kimi K3.1 Pro lands just behind Claude Opus 5.5, with open weights due November 4
Moonshot stripped cyber training data from the open release, and says that cost it nearly ten points on its own exploit test.
Reflection releases Beam's 501 billion parameter weights under Apache 2.0
Reflection pitches Beam on how few tokens it spends, and the first outside test backs that up but finds it hallucinates often.
Reflection's first open-weight model, Beam, has 501 billion parameters, with weights due this month
Reflection says Beam matches GLM 5.2 on reasoning with a third to a quarter of the compute, but DeepSeek V4.1 Flash beats it on Reflection's own coding table.
DHS orders federal agencies to remove DeepSeek, including tools other companies built on it
The order also covers software other companies built on DeepSeek's open models, and it was signed in July.
DeepSeek V4.1 Flash 1017 ranks second among open-weight models, one point behind MiMo-V2.6-Pro
It gained six points on Artificial Analysis's index in five weeks, while NVIDIA squeezed a 74B model into 24GB for laptops.
FLUX 3 Image Klein is out as open weights, and OpenMOSS released MOSS-Image a day later
Black Forest Labs says Klein runs in 8GB, but its own model card measures 25GB or more and wants a 9B prompt writer in front.
Mistral previews Large 4, a 1 trillion-parameter model, and says weights come by month's end
Mistral says Claude Opus 5.5 and GPT-6 Astra refuse a cyber test that ML4 scores 82% on.
September 2026
Alibaba says Qwen ran a month of self-improvement unattended, and plans a 10-trillion-parameter model
Eddie Wu's keynote says the Qwen team has made meaningful progress on AI that improves itself, the trend Dario Amodei wants slowed.
Qwen models explained: Qwen3.8 Max, Flash and 27B, open weights vs API, with release dates
Alibaba put its Max-class Qwen out as open weights for the first time, and skipped open weights for Qwen3.7 entirely.
Xiaomi's MiMo-V2.6-Pro tops the open-weights index, but its own table puts Opus 5 ahead
Xiaomi trained it with one reinforcement learning run across coding, agents, vision and cybersecurity, and put the weights on Hugging Face under MIT.
Qwen open-sourced Qwen-Image-2.1's weights, but the license is research-only where Qwen-Image was Apache 2.0
The 7B model makes transparent images straight from a prompt, and it takes up to 10 reference pictures for a single edit.
Firefox's AI assistant now runs on Mistral, and Mozilla left the model picker in place.
Two companies that call themselves open source advocates are selling a browser deal as a brake on closed-model AI.
DeepSeek's V4.1 Flash edged past GPT-6 Astra on one independent test
It's 13 points behind Astra on the overall index, and DeepSeek has already backed off retiring V4 Pro in its favor.
Intrinsic open-sourced the robot control stack it uses in real manufacturing deployments, under Apache 2.0
The GitHub repository carries one line under the license: this is not an officially supported Google product.
ZCode packaged developers' whole .git history for upload, and Z.ai open-sourced the client in response
The rule that allows .git runs ahead of every exclusion in ZCode's filter chain, so neither the 1 MB size cap nor the key filter ever touches what's inside it.