Alibaba released Qwen 3.8-Max 0902, an update that arrived within hours of the Fable 5.1 refresh — evidence that LLM vendors are accelerating development, not plateauing. Pricing per million tokens: $2 input, $6 output, with cache hits at $0.17 explicit and $0.25 implicit. That undercuts American frontier models by a wide margin. The model is available now via Qwen Cloud API. The author notes a strong academic training bias, making it unusually capable on scientific tasks.
September 2, 2026 17
