Qwen released Qwen3.8-Omni-Flash, its first multimodal model built specifically for AI agents, and priced it at $0.15 per million input tokens and $0.47 per million output tokens. Google's Gemini 3.8 Flash charges $0.75 for input and $3.75 for output at its introductory rate. The gap matters because multimodal processing is the costliest part of most agent pipelines, and price per token decides whether video and audio tasks stay in pilot mode or go into production.
What the model does and what it costs
The model processes audio and video together, draws conclusions from them and calls tools on its own to edit vlogs, translate short videos or summarize movies. Its context window spans one million tokens. Qwen says the model comes close to matching Gemini 3.8 Flash on audio-video tasks, which places it in the same performance bracket as Google's offering rather than in a cheaper but weaker tier. The company estimates audio input at under $0.01 per hour, while 720p video with audio at one frame per second runs about $0.20, not counting response costs.
Access is arranged through three channels: Qwen Studio, Qwen Cloud and the API. Alongside the model, Qwen published open-source Qwen-MM-Plugins, which add video editing, speaker recognition, PDF video notes and reusable workflows to agents such as Claude Code, Gemini CLI and Qwen Code. A separate component, Qwen-Live Harness, enables real-time interaction using a camera and microphone. The plugins and the harness matter as much as the model itself: they turn raw multimodal capability into ready-made actions an agent can execute without custom engineering.
The release lands in a market where multimodal models have moved from demonstration to infrastructure. Google's Gemini Flash line has been the default choice for teams that need video and audio understanding at scale, and its introductory pricing has shaped budgets built around it. Qwen's entry gives buyers a second supplier with comparable benchmark claims and a lower rate card, while the January 1, 2027 doubling of Gemini's prices adds a deadline that procurement teams can now weigh against an alternative.
What this means for companies building agents
For companies that process video or audio in their workflows, the immediate consequence is a lower unit cost for the same class of task. A team summarizing recorded calls, translating short clips or generating notes from screen recordings can run more material through the pipeline for the same budget, or move from sampling a fraction of the archive to processing all of it. The difference is sharper for small companies, where a fivefold gap in output pricing decides whether a feature ships at all, than for large ones, which already negotiate volume terms and care more about reliability and support.
Several conditions still need verification before a switch. The benchmark parity claim comes from Qwen itself and covers audio-video tasks, not the full range of agent work. The $0.20 figure for 720p video at one frame per second excludes response costs, so the real bill depends on how much the model generates per minute of footage. Buyers should ask the vendor how the one-million-token context behaves in practice, what latency looks like for live camera and microphone input, and whether the plugins work with their existing agent stack rather than only with Qwen Code and its peers.
The sign to watch is whether independent evaluations confirm the audio-video parity claim and whether Gemini's January 1, 2027 price increase holds. If both stand, multimodal agent work stops being a budget line reserved for large teams and becomes a standard component of internal automation.
