Thud Audio Action Model Claims 10x Faster, 20x Cheaper Than Gemini Live
AnhPhuNguyen1 · x · 2026-10-15
Caine Ardayfio and AnhPhu Nguyen unveiled Thud, an "Audio Action" model that skips transcription entirely — the AI listens and acts directly.
Thud works against a pre-built list of actions available on each page and matches spoken commands to those actions. Combined with Mobile Use and Jev for audio, the team claims it runs 10x faster and 20x cheaper than Gemini Live. Their thesis: in 2026, models are smart enough to act on raw audio without a text intermediate step.
Related event: Thud Audio Action Model Responds in 200ms, 10x Faster Than Gemini Live(2 posts)→
More from coding & agent
- Yacine: AI agents shouldn't write 100k lines of code — more code is a liability — yacineMTB · 2026-10-15
- Claude Code Spins Up 161 Codex Sessions to Recreate a 63-Second Video — ZacharyHuang12 · 2026-10-15
- Running Claude, Codex, Gemini and Grok in parallel: 1,708 queued tasks, the human is the bottleneck — nptacek · 2026-10-15
- Claude Desktop Gets Minimalist: Subtasks Shown, Thinking Hidden, Intervention via Effort — aigclink · 2026-10-15
- Yacine: Every LLM-written codebase I've reviewed had something nefarious snuck in — yacineMTB · 2026-10-15
- Why Build Agent Plugins at All? Devs See Little Durable Value Without Network Effects — heyneighbor · 2026-10-15