You wouldn't wear a scuba suit to a sprint. So why use one speech-to-text model for every job?
AssemblyAI-STT-3.7 is three models: pre-recorded, Realtime, and Realtime Agent.
assemblyai.com/blog/introduc…
1
1
368
AssemblyAI retweeted
Today we're launching AssemblyAI-STT-3.7 - three new transcription models with industry-leading accuracy on voice agent calls, live conversations, and recorded audio.
We rewrote our entire inference engine for these models, so they're now even faster and more scalable.
All three models are live now:
AssemblyAI-STT-3.7-Realtime-Agent:
- Perfect for voice agents
- #1 in accuracy for noisy telephony voice agent calls
- 86ms to the final transcript, 2x faster than our prior model
- 32 languages
AssemblyAI-STT-3.7-Realtime:
- Perfect for live meetings and captions
- The lowest speaker diarization error rate on live calls and meetings
- 32 languages in one stream with automatic language detection
AssemblyAI-STT-3.7 for recorded audio:
- Leading accuracy across 32 languages
- The lowest speaker diarization error rate of any pre-recorded STT API
- A 5-minute file comes back in 9 seconds
All available today in our API.
Learn more here: assemblyai.com/blog/introduc…
Introducing AssemblyAI-STT-3.7: our next generation of transcription models.
Live conversations, voice agents and recordings fail in different ways. A meeting transcript that can’t tell two speakers apart needs speaker labels that are right the first time, not fixed later. An agent that hears a “no” too late needs the final transcript before the caller notices the pause. A recording where the right words land on the wrong person needs accuracy and attribution scored together. Those are three different problems, so AssemblyAI-STT-3.7 ships as three models, each built around the one that matters for its audio.
-> AssemblyAI-STT-3.7-Realtime for live conversation. Final speaker labels the moment a turn ends, the lowest live diarization error on calls and meetings, and pricing that makes real-time a real alternative to async.
-> AssemblyAI-STT-3.7-Realtime-Agent for voice agents. Final transcripts in 86 ms, with the lowest word and entity error on voice-agent calls, even over background speech.
-> AssemblyAI-STT-3.7 for recorded audio. 32 languages and the lowest speaker-attributed error rate of any pre-recorded API.
All three are available today.
Read the full announcement for benchmarks and side-by-side recordings against other providers, or test it on your own audio in the playground: assemblyai.com/blog/introduc…
Accurate, fast and affordable speech-to-text is coming to cars 🤯
We put Blurt on Android Auto. Its built-in voice input heard "take the swimming complex." Blurt heard Teck Ghee.
assemblyai.com/blurt?utm_med…
AssemblyAI-TTS-1 is the 5th-fastest model on Artificial Analysis's new Time to First Audio benchmark: 229 ms, averaged over 7 days of daily runs, silence included.
That's the gap a caller actually hears. Built for live calls.
Announcing Artificial Analysis Time to First Audio, measuring how quickly Text to Speech models start speaking. Among the leading models the wait ranges from 76 ms to nearly a second
Our Text to Speech benchmarks already cover quality, pronunciation and generation speed, but not how long a user waits before hearing anything — the number that decides whether a voice agent feels responsive. We measure every model daily on the same 12 prompts across four input lengths, timing audible speech rather than the first byte, and report a rolling 7-day mean.
Key results:
➤ Fastest at the frontier: @SpeechifyAI's Simba 3.5 starts in 116 ms — the highest Arena Elo of any model that begins speaking in under 200 ms. @InworldAI's Realtime TTS-2 follows at 186 ms.
➤ Turbo variants earn their name: @ElevenLabs' Eleven v4 Turbo starts in 258 ms against 653 ms for Eleven v4, at the same Arena Elo. @FishAudio's Fish Audio Drama 3 matches it at 258 ms and @cartesia's Sonic 3.6 is at 287 ms.
➤ Streaming is the dividing line: The 65 streaming models typically begin speaking within a few hundred milliseconds, a median of 376 ms. The 20 non-streaming models must finish generating before they return anything: a median of 2.0 seconds.
See more details below ⬇️
We opened an 8 Ball and put text-to-speech in it to see what would happen. Follow to see the full build on Monday!
🆕 GPT-6.1 Terra, Beam-1, DeepSeek V4.1 Flash 1017, and Muse Spark Open are now available on LLM Gateway!
See which model is right for you: assemblyai.com/products/llm-…
Who is Medical Mode built for? We hopped on a huddle with Britney, one of our PMs.
Explore AssemblyAI for Healthcare → assemblyai.com/solutions/hea…
3
9
880
congrats to the team at @retellai 👏
Retell just crossed $100M in revenue run rate.
$10M to $100M in 17 months. Profitable. $4.6M raised.
From day one, we’ve focused on solving the hardest parts of deploying voice AI in the product itself, so customers can build, manage, and improve their own agents without depending on us.
Our bet: the future of the contact center belongs to the teams who run it.
Proud of our team. Grateful to every customer who believed in us.
20
1,351
WE ❤️ PETS HERE AT ASSEMBLYAI, which is why we just released Vet Mode transcription
Check it out here --> assemblyai.com/solutions/vet…
Can Medical Mode transcribe 23 of these medical terms right? Read rapid fire into the AssemblyAI playground. Watch the counter.
assemblyai.com/solutions/hea…
🐶 Vet Mode is tuned to correct veterinary drug names that general models often replace with human ones, like hearing 'ibuprofen' but writing 'carprofen'.
1
1
96
Healthcare teams need the most accurate voice AI for their clinical products, which is why they build on AssemblyAI.
See what's new: assemblyai.com/blog/assembly…
1
85
Today, we’re releasing two updates that make healthcare accuracy even stronger: an updated Medical Mode for pre-recorded and realtime audio, and Vet Mode.
A clinical note can be 99% accurate and still get a drug name wrong. That last 1% is the part a clinician has to catch 🧵
2 of my 6 tips for getting into hardware hacking, from one beginner to another. All 6 in the full video: youtube.com/watch?v=uY4-ZMDV…
AssemblyAI retweeted
Hosted my first every Hardware hackathon at @AssemblyAI
can't wait to see you all for the next one, sign up below!
We built Blurt for Google Forms. Now you can just dictate your answers and have your Google Form filled in automatically.
Blurt is free and open source: assemblyai.com/blurt?utm_med…








