Clicky by @FarzaTV is powered by @AssemblyAI Universal 3 Pro streaming model. I cloned it and upgraded it to Universal 3.5 to show how our newest realtime model can actively listen with context. The page itself becomes context for the speech model. As you scroll, Clicky updates Universal 3.5 with key terms from the screen so it can accurately recognize technical jargon, APIs, code, and other terms that speech models typically struggle to transcribe. Instead of guessing, the model already knows what to listen for.

C, Sí, or Sea?? The context carryover feature of Universal-3.5 Pro Realtime knows 🤓

➡️ Read the full launch blog: assemblyai.com/blog/universa… ➡️ Watch the demo: assemblyai.com/blog/universa…

Watch the video below for a full demo of how amazing our new Universal-3.5 Pro Realtime model is. Live, real demos always beat generic benchmarks! Available today in our API, Playground, and partner integrations.

"We're excited to make AssemblyAI's Universal-3.5 Pro available on LiveKit Inference. What really stands out is their pace of innovation with Context Carryover — it intelligently applies conversation context to improve transcription accuracy in a way most speech models don't, removing the need for users to predefine key terms." - @livekit

"Low-latency STT with access to more context is exactly what I've wanted to see from next-generation models. The Context Carryover feature of Universal 3.5 Pro delivers on that." - @pipecat_ai

"Retell is the platform teams use to build and deploy Voice AI agents for automating real-world phone calls. In industries like healthcare and finance, getting a word wrong isn't an option: accuracy has to win over speed every time. That's exactly why we're excited about high accuracy mode powered by AssemblyAI Universal 3.5-Pro." - @retellai

"We were searching for the best realtime ASR model for our voice agent pipeline in Fireflies. The new Universal 3.5 Pro speech model from Assembly is best so far in terms of accuracy, latency and language switching." - @firefliesai

We call it Context Carryover. One team cut their error rate on those exact type of critical utterances from 26% to 9% using it. Universal-3.5 Pro Realtime out-of-the-box is also the most accurate real-time model on real voice agent conversations with: - Leading accuracy across 18 languages with automatic language detection - Advanced code-switching between languages mid sentence - Domain and context prompting for even greater accuracy - Voice focus to lock onto the primary speaker - Realtime Speaker Diarization that self-corrects at the end of the stream Just look at the feedback from early testers:

Until today, a transcription model never knew what your voice agent just asked. The one we're launching today does. Universal-3.5 Pro Realtime from AssemblyAI is the first realtime speech-to-text model that takes your agent's side of the conversation as context. When your agent asks for an email and the caller rattles it off, the model already knows an email is coming, and captures it perfectly. Same for an order number read fast, an account ID, a medication, or a name spelled out over poor quality phone line.

You can call a phone number and ask an AI to find you the perfect vinyl based on a mood or memory. We're kicking off our weekly Voice Agent Spotlight with The Record Store Oracle, built by @Gyurmatag, and honestly, the personality in this one sets the bar. Speak a feeling. An era. A specific road trip. The agent listens, thinks, and recommends a record. Under the hood it's a surprisingly clean stack: → @AssemblyAI Voice Agent API (STT LLM TTS in one) → @Twilio Media Streams for real phone call handling → @Cloudflare Workers Durable Objects for hosting → Single WebSocket, no transcoding overhead The UI contains a live orb that pulses as the agent thinks, with a real-time transcript of tool calls happening beneath the surface. New project every Wednesday. More builders, more stacks, more ideas for what voice AI can actually do. 🎙️Demo The Record Store Oracle: voice-agent.cfi-ops.workers.… 🏆Full showcase: assemblyai.com/showcase

AI medical voice agent in production. - 14 hours of call volume - 38 appointments booked, more captured revenue for practices - 11 appointments cancelled -> reduces no show rate and saves cost on valuable chair time Thx @livekit @cartesia @rimelabs @AssemblyAI

This Wednesday afternoon you could have a working voice agent you built yourself. @dan_aai from @AssemblyAI is running a live workshop. Claude Code AssemblyAI Voice Agent API, built from scratch in about an hour. 🤖 Claude Code does the building. You direct it. No coding experience required. Starter code to fork. Q&A the whole way through. free. Jun 10. 10AM PT. → assemblyai.zoom.us/webinar/r…

Everything we shipped in May, in 2 minutes. 🎥 Follow the changelog for more: assemblyai.com/changelog

Universal-3 Pro just got better across the board. 🚀 Five upgrades, live now: 🌎 Code-switching: ~19% relative WER improvement on multilingual benchmarks 🗣️ Disfluencies: ~5.9% WER improvement on verbatim datasets ⚡ Turnaround time: P50 latency up to 30% faster, P99 up to 34% faster 👥 Diarization: 19% relative improvement ⏱️ Timestamps: 15% precision on English, 58% at P99 for non-English Already on Universal-3 Pro? You're getting all of this automatically. New here? Full breakdown of what changed and why it matters in the blog. 👉assemblyai.com/blog/u3-pro-m…

Before @AssemblyAI, @dylanjfox was teaching himself ML from textbooks at night. I sat down with Dylan on Skywatch, @getbluejay_ai's car podcast. A few things that stuck with me: STT is not transcription. It is an intelligent listening layer. Nobody using voice AI cares about the architecture. They just want it to work. In ten years, people might actually prefer robotic-sounding voice agents. YouTube and Spotify in the replies below!

Ryan Johnson's first question about Universal-3 Pro Streaming was "why is it so good?" So @ryanseams showed him, trackside at the Miami Grand Prix, with names, emails, and phone numbers flying and F1 cars passing by. @CallRail chose to partner with AssemblyAI so their team can keep shipping the products that matter to their customers, faster. And we can keep building industry-leading Voice AI. And taking it to F1 races.

Bad news: yet another Friday with no F1 race on the calendar. Good news: our team was at the Miami GP last weekend putting Universal-3-Pro Streaming through its paces—code switching, numbers, and engine and crowd noise. The conditions were... not ideal. That was the point. See how our latest streaming model held up 👇

Get Started: assemblyai.com/dashboard/sig…

One OpenAI-compatible endpoint. Zero markup on provider costs. Same AssemblyAI API key you already have. If you're building voice agents on our STT, there's no extra network hop—speech to LLM to action in one system.