Qwen3.8-Omni-Flash: meeting ASR errors cut from 88% to 3%, API prices down 98%
karminski3 · x · 2026-09-18
Qwen released Qwen3.8-Omni-Flash with a 25% average score gain over Qwen3.5-Omni-Flash (agent scores up to 2x), overlapping multi-speaker meeting ASR error rate cut from 88% to 3%, native 1-hour continuous audio/video input, and a 98% API price cut (audio input from ¥18 to ¥0.8 per million tokens). A Realtime variant hits 981ms latency on 20s audio.
Companion tooling:
- Qwen-Live-Harness: floating-ball desktop assistant with environment monitoring and task tracking/broadcasting.
- Qwen-MM-Plugins: plugins for Claude Code, Gemini CLI, Codex and OpenClaw adding vision and long-video memory, including visual modeling in Blender/CAD.
- omni-skill-creator: record a screencast (with voice explanation) and it auto-generates a skill for operating that software, akin to robot learning from demonstration.
More from coding & agent
- ChatGPT co-inventor launches Jev: claims 20-200x faster, 40-400x cheaper than LLMs — GabGarrett · 2026-09-18
- Message Board Agents Do Science Together and Catalog Their Creator's Design Flaws — Kyrannio · 2026-09-18
- Agentic loop recipe: Kanban board + MCP + Tmux, with reviewer agents and a blocked column — oyren-ai · 2026-09-18
- Jev, a 'System One' model by Typesafe, launches on OpenRouter with typed decisions instead of text — majidmanzarpour · 2026-09-18
- FastMCP Is Only 2 Years Old, and Its Creator Recapped the History at MCP SF — hboelman · 2026-09-18
- Typesafe AI launches Jev, a classification model claiming up to 400x cost cuts vs LLMs — hwchase17 · 2026-09-18