Qwen3.8-27B Now Runs at Cerebras Speed
- Source
- x.com
- Date
Alibaba's Qwen team announced that its dense Qwen3.8-27B model is now served on Cerebras hardware, giving it the fast inference speeds Cerebras chips are known for instead of standard GPU serving. Artificial Analysis scores the model close to GPT-5.6, DeepSeek V4 Pro, and Claude Sonnet 4.6 on general intelligence benchmarks, a strong result for an open weight model at this parameter count. Early testers, however, report that Qwen3.8-27B underperforms specifically on coding tasks relative to its general benchmark score, a reminder that a single intelligence index number can mask weakness in a domain that matters most to a given team. For developers who need fast, cheap inference on non coding workloads such as classification, extraction, or general reasoning, the Cerebras backed Qwen3.8-27B is now a legitimate open alternative to closed frontier models. Anyone evaluating it for coding specifically should verify against their own tasks rather than relying on the aggregate score.
Fast meets open. 🚀 Qwen3.8-27B is now running on @cerebras with rapid inference. Try it now! https://t.co/iBV1v8Ix7d
Context
Alibaba's Qwen team says its dense, Qwen3.8-27B model is now available through Cerebras, a chip vendor known for serving models at unusually low latency because of very high on-chip memory bandwidth. The post doesn't publish a specific speed or cost number for this deployment, so the concrete size of the gain over running the same model elsewhere isn't established by this announcement alone.
- open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
- inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
Checking sign-in…
Loading comments…