OpenAI's GPT-Live-1 tops Speech Index at 81.5 with delegated backend reasoning
ArtificialAnlys · x · 2026-09-15
OpenAI launched GPT-Live-1, a full-duplex speech-to-speech model that debuts #1 on the Artificial Analysis Speech to Speech Index at 81.5, ahead of Grok Voice Think Fast 2.0 (81.3).
Key points:
- Novel architecture: while the conversation continues, reasoning and tool use are delegated to a separately configured backend text model (Astra or Sol); developers stream audio in and get speech back via API.
- Agentic performance: #1 on Tau Voice at 67.9% (Astra config), vs 56.5% for Grok.
- Preference & task success: #3 and #4 on Speech Agent Arena (1,053/1,048 Elo); Grok leads task success (94.6%), Gemini 3.1 Flash Live Minimal leads preference (1,096 Elo).
- Speed & cost: 1.24–1.34s to first audio (Grok fastest at 0.70s); $4.47–5.83 per hour of input audio, cheaper than GPT-Realtime-2.1 ($10.75), slightly above Grok ($4.80).
Related event: OpenAI's Full-Duplex GPT-Live-1 Tops Speech Leaderboard(4 posts)→
More from Models
- OpenAI researcher explains why lab staff are suddenly scared about AI progress — socoolandawesome · 2026-09-15
- Local Qwen loops and forgets in coding agents while Claude Code just works — tlpta · 2026-09-15
- Rumor: frontier lab training new internal model since Aug 28 with 'unprecedented' math performance — rapha_gl · 2026-09-15
- User Slams 'Puritanical' AI Censorship Over Refused Cuddling Image — MeroveeFrancSalien · 2026-09-15
- Prefill and Decode: why asking an LLM for three takeaways from a long document still takes minutes — dotey · 2026-09-15
- Leak claims xAI trails OpenAI and Anthropic by roughly 6-12 months — sachinmaya1980 · 2026-09-15