The introductory rate for Google's Gemini 3.8 Flash sits at $0.75 per million input tokens and $3.75 per million output tokens. Qwen has now released a multimodal model priced at $0.15 and $0.47 respectively.
The model, Qwen3.8-Omni-Flash, is the first multimodal offering Qwen has designed specifically for AI agents, and the company claims its results on audio-video tasks land close to those of Gemini 3.8 Flash. The selling point is straightforward: nearly equivalent performance for a small slice of the cost.
How the cost breaks down on an actual task
According to Qwen, audio input runs at less than $0.01 per hour. A 720p video with audio, sampled at one frame per second, comes out to roughly $0.20 by the company’s estimate, although that number leaves out the cost of the generated response.
Gemini’s pricing is heading in the opposite direction. Google has scheduled its introductory rate to double on January 1, 2027, meaning the gap Qwen is promoting now would grow further should Google stick to that plan.
Designed for agents rather than simple prompting
The model processes audio and video together, reasons over both, and calls tools without being told to. Qwen cites vlog editing, short-video translation and movie summarization as example use cases. Its context window reaches one million tokens.
I have not yet tested it under a production workload, so the benchmark claim belongs to Qwen rather than to me. Note also that the company’s own wording is that it comes close to matching Gemini 3.8 Flash, a qualifier worth reading carefully before moving any workload over.
Availability
Qwen3.8-Omni-Flash can be accessed via Qwen Studio, Qwen Cloud and the API. The open-source Qwen-MM-Plugins add video editing, speaker recognition, PDF video notes and reusable workflows to agents such as Claude Code, Gemini CLI and Qwen Code. For real-time interaction over a camera and microphone, there is Qwen-Live Harness.
Anyone currently paying Gemini 3.8 Flash’s output rate for video summaries should start by looking at that $0.47 figure.