Gemini 3.8 Flash

Alibaba released Qwen3.8-Omni-Flash on 18 September 2026: an omni-modal model built on the Qwen3.8-Flash-Next architecture that accepts text, images, audio and video into a 1 million token context window and returns text. The headline is price — more than 98% off an hour of audio input against the previous generation — but the mechanism is more interesting. Agentic perception lets the model decide what to watch and listen to, raising OmniVideoBench accuracy from 63.4 to 67.8 while cutting tokens per query by about 45.7%. This article covers the specification, the benchmark claims, the pricing arithmetic, the Apache-2.0 plugin suite and the limits.