Vapi

Today | 10:00 AM PDT A scripted chatbot handles the easy calls. Nikhil Gupta, Co-Founder and CTO of Vapi, and Marius Buleandra, Member of Technical Staff at Anthropic, get into what frontier intelligence and a voice platform do that a script can't. Real deployments, a live example, and audience Q&A, in 45 minutes. Join here → https://lnkd.in/gh5vHS-B

We're #hiring a new Billing and Revenue Accounting Specialist in San Francisco, California. Apply today or share this post with your network.

One week from today! Join us for an evening of boba, Korean bites, and Durable AI demos with our friends at Temporal Technologies , HeyGen , and Modal You'll hear from Adham El Fawal about building a durable system for scaling outbound campaigns with voice agents whether you are making 1 call or 10,000! RSVP here: https://lnkd.in/dVJ7ETuZ

The #1 question we get from builders: "Which models should I use?" A voice agent relies on a transcriber, an LLM, and a voice, picked from a model catalog that advances almost every day. Today we're launching Vapi Model Intelligence, a bundle of two features, to answer it. This includes: Model Presets: curated configurations tuned to a goal. Balanced, High Intelligence, Ultra Fast, and Cost Saver each bundle a best-in-class transcriber, LLM, and voice. Pick one and ship. Model Performance Metrics: current cost, latency, and quality data on every model, drawn from real Vapi deployments, including proprietary metrics like our Humanness Index™ for voices. Compare models on the metrics that matter most for your use case. Ship strong by default. Optimize when you're ready. Read more: https://lnkd.in/eiT8j78A

We're #hiring a new Cost Revenue Analyst in San Francisco, California. Apply today or share this post with your network.

Voice AI Live: Using Deepgram Models in Vapi Same models, very different jobs. Amanda and Ed talk choosing Deepgram's Voice AI models Nova-3, Flux, and Aura-2 on Vapi and where each earns its place. English vs multilingual, medical and custom. Topics include transcription vs real-time turn detection and latency concerns.

Chronic disease management programs are effective when patients respond to passive channels, such as text, email, or chat. But what about hard-to-reach patients? How do you re-engage when a monitoring device stops responding, a refill sits idle at the pharmacy, or a patient goes quiet? Signals like these only detect disengagement; they can’t tell you why. Voice can. A live call from a Vapi voice agent uncovers why the device isn't working and walks the patient through fixing it so that readings can resume. The one-pager shows 3 key workflows where voice AI re-engages patients who’ve gone quiet, powered with full context from your RPM, care management CRM, and EHR. Download the doc

We're #hiring a new Product Designer — AI UX / Product Experience in San Francisco, California. Apply today or share this post with your network.

Scripts are fine until a caller says, "Actually, cancel that, and can you check my last order too?" before the agent finishes the first request. A scripted flow stalls, or routes to a person who starts over. A model like Claude works through what the caller actually said and keeps the call moving. With a platform like Vapi, you still set the boundaries: what the agent can do, when it hands off. Inside them, the conversation adapts rather than breaking entirely. On July 22, Vapi Co-Founder and CTO Nikhil Gupta sits down with Marius Buleandra, Member of Technical Staff at Anthropic, to talk through what a voice platform plus frontier intelligence unlocks, from high-volume inbound support to complex multi-step work, with real deployment examples and a live demo. Wednesday, July 22 | 10:00 AM PDT Registration link in comments.

The best voice models now sound nearly human. Three weeks ago, we launched the Humanness Index™, a blind, crowdsourced leaderboard that answers the question voice benchmarking always skipped: does this voice model sound like a human? In every battle, two models read the same quote in the same cloned voice, with a real human recording included in the comparison set. So the scores mean something. After 11,000+ votes, xAI's Grok TTS leads at 95, within five points of the real human at 100. Many listeners struggle to tell them apart. Why it matters: for years, the ceiling on voice automation wasn't just intelligence; it was the voice itself. Callers never experience an agent's reasoning directly. They experience the voice that carries it. And the voice is finally keeping pace. The top four models are within a few points of each other, and every vote sharpens the picture. Cast a vote: humannessindex.vapi.ai Link to full results in comments