Alibaba's new 27B parameter multimodal dense model is now available under an Apache 2.0 license with a native context window of 262K tokens, extendable to 1M via YaRN. The Qwen3.8-27B model ranks 9th overall in the Code Arena WebDev benchmark with 1,595 points, making it the only model in its size class to enter the top 10. It sits six ranks behind the larger Qwen3.8-Max and ahead of Google's Gemma 4-31B, which ranks 80th.
The model is available on Hugging Face and ModelScope, while Venice offers a private and uncensored version for agent deployment. Optima has introduced benchmarking for the model to evaluate speed and cost on local laptop hardware. Alibaba also released open weights for the 2.4T parameter Qwen3.8-2.4T-A95B Max-level model to support high-capacity agent construction.
Key sources
- SOURCE@arena“Gemma 4-31B which was released back in April sits at #80”x.com
- SUPPORT@arena“#6 Consumer Product (1587) - #7 Brand & Marketing (1627) - #8 Gaming (1645)”x.com
- SUPPORT@artificialanlys“See how a model that can run on your laptop compares on performance, cost and speed for your custom use cases”x.com
- SOURCE@arena“262K native context, easily extendable to 1M tokens via YaRN”x.com
- SUPPORT@arena“shifted the Pareto frontier for Image-to-WebDev Arena!”x.com
- SOURCEhuggingnewshuggingnews.com