Ant International's Heartbeat Thinking cuts voice-agent latency by 36.4% with speculative tool calls
ant-intl · hf · 2026-10-23
Ant International published a paper introducing Heartbeat Thinking, a speculative reasoning scheme for realtime voice agents. Instead of waiting for speech to end (or truncating tool calls on every update), ongoing invocations are allowed to run to completion while periodic 'heartbeats' read the latest partial transcript and a shared record of completed reasoning/tool results to fill in missing arguments and launch dependent calls. Across SpeechFCEval, FDBench v3 and AudioCRAG it beats SHANKS-style windowed baselines; on FDBench v3 it raises tool coverage by 16 percentage points and cuts mean final-response latency by 36.4% without hurting task quality. Paper is on arXiv (cs.CL).
More from coding & agent
- Dev Jokes: You Can Only Relax Using Claude on Stuff That Doesn't Matter — louisvarge · 2026-10-23
- PHP Has Been 'Dead' for 30 Years — and It's Still Not Dead — 0xkarasy · 2026-10-23
- VC: By 2027 companies won't be able to justify hiring non-agent engineers — latticecut · 2026-10-23
- Opus 5.5 codes an entire sci-fi movie title sequence—every frame and synthesized note — prasenx · 2026-10-23
- Local AI slide generation for privacy-constrained healthcare: HTML vs PPTX — AdRepulsive7837 · 2026-10-23
- MiniMax Design MCP goes live: create with Design inside Codex and other agents — Hailuo_AI · 2026-10-23