Artificial Analysis releases full streaming STT leaderboard and methodology
ArtificialAnlys · x · 2026-10-02
Artificial Analysis published full results and methodology for its streaming speech-to-text benchmark. The AA-WER Streaming index uses 8 hours of audio weighted across AA-AgentTalk (50%), VoxPopuli (25%) and Earnings22 (25%), covering diverse accents, domain language and tough acoustics, comparing 37+ models on WER, latency after end of speech and price, with a Pareto-frontier view split by proprietary vs open weights.
More from Models
- Bug Hunt Benchmark: Opus 5.5 nears Fable 5.1 at 2/3 cost; GPT-6.1 Sol 10x cheaper — PawelHuryn · 2026-10-02
- Kevin Kern: new GPT looks great, but right-click new chat is missing — kevinkern · 2026-10-02
- Grok down for some users, responses delayed by minutes; xAI says it's on it — Daniel_Farinax · 2026-10-02
- Subscription vs API pricing tested: Claude Max 20x gives double OpenAI's effective subsidy — PawelHuryn · 2026-10-02
- Google restricts Gemini 4 Argon to vetted cybersecurity experts over hacking misuse fears — nordicinst · 2026-10-02
- Claude Sonnet 5.5 xHigh lands #3 on Code Arena WebDev, 2 pts behind GPT-6 Astra — arena · 2026-10-02