🩺 @sullyai has returned over 30 million minutes to physicians — more time with patients, less on paperwork. @baseten powers this with their optimized inference stack built using NVIDIA Blackwell, NVFP4, TensorRT LLM, and NVIDIA Dynamo, to run frontier open models like gpt oss 120b. The result: 10x cost reduction and 65% faster responses for workflows like clinical note generation. 🔗 Read the blog: nvda.ws/468smA3

We gave Jev 2,029 real phone calls. No transcripts or audio; it never heard a word. Our AI receptionist's calls were reduced to pure structure, meaning turns, tool calls, workflow stages and timing. During the calls, Jev made 38,012 turn-level forecasts at 118 ms median latency, reviewed every call with five typed questions and produced 10,145 answers in 26 seconds with 256 requests in flight. The experiment was zero-shot, with no fine-tuning or examples from our data. We compared Jev's forecasts with what actually happened in the EHR. By the halfway point, Jev could meaningfully separate calls that would book from those that wouldn't (AUC 0.78), and near the end it ranked them correctly 94% of the time. Even though Jev over-focused on visible errors our agent usually overcomes, it's still pretty incredible that it analyzed thousands of real calls in seconds for only $3.

It's actually weird feeling to watch recursive self-improvement start working in your own harness, and to finally see the results. With Opus 5.5 and GPT-6 Astra, our harness now improves itself and agents do almost every step of it. 1. Agents write the evals. They turn failing call traces into versioned scenarios: caller persona, goal, the tools that should run, and the final state expected in the EHR. 2. Agents call agents. Simulated callers are wired to phone our deployed receptionist in real rooms, with real tools and a sandbox EHR, in text and audio, across both of our runtimes. 3. Agents grade agents. An LLM judge plus deterministic end-state checks score each call, and failures are clustered by runtime, tool and check. Agents then write one fix and one new scenario per root cause as a stacked PR. 4. Agents review agents. Reviewer agents critique every PR, findings get fixed with tests, and scenarios are rerun. A change proposed only if it passes on both runtimes. We still own merge and deploy, for now. We're early, but for the first time the loop is starting to close. I've never experienced a pace like this. It's getting faster so RAPIDLY.

Sully signed the Open Weights letter today. Most of our clinical stack now runs on open models and we're migrating the rest. Inference costs dropped 90% and latency dropped 65%. But cost and speed aren't the reason we keep going. Open weights are what let us post-train models for the specialties our clinicians actually practice in, run an ensemble of specialist models that check each other rather than trusting one general model, and keep patient data on infrastructure we control. None of that is possible when the weights sit behind someone else's API. A better open source model shows up every few weeks, and we can have it in production within days of release. That keeps pushing our costs down, which changes who we can afford to serve. A rural clinic with four doctors should get the same AI staff as a large hospital system. I'm proud of the team that did the work. The infrastructure partnerships, the migration, the long hours behind both. That's what put us on this list, and it's what has us ready for what comes next.

Kimi K3 may be an important inflection point for AI. Potentially negative for Anthropic and OpenAI while being net positive for essentially every other company in the world. I mean that very literally. Although the real “Sputnik moment” would be an open-source frontier model that was also token efficient unlike Kimi K3 which is 50-70% more expensive to run than GPT 5.6 per Artificial Analysis. Rationale:   A world where there are only 2-3 dominant frontier labs with 90% inference margins is net negative for every other layer while being awesome for those 2-3 labs. Those labs would become monopsonies for power, data centers, semiconductors and hyperscalers and would obviously vertically integrate over time into all those layers while also completely subsuming the application/software layers.    Anything that lowers margins and increases competition at the model layer is good for every other AI layer: power, semiconductors, hyperscalers, neoclouds and yes even software.   This is why Jensen is so supportive of open-source. An open-source model requires the *exact* same amount of compute to run as a closed frontier model of similar size and architecture. Kimi K3 is roughly the same price as GPT 5.6 Terra on a per token basis, which actually suggests that it is less computationally efficient as I am sure that GPT 5.6 is priced to a higher margin than K3. And given that K3 is a token wastrel, i.e. token inefficient, it is significantly more expensive per task than GPT 5.6 and Grok 4.5, which are much more token efficient. Cost per token and token efficiency (i.e. intelligence density per token) are the drivers of intelligence per unit of cost. The winning AI companies will be those that offer the most intelligence per $ over time.   Lower margin % at the model layer = more margin $ at every part of the infrastructure layer and is a godsend for software. This can happen either through open-source models like K3 at the frontier *or* having a vertically integrated model company like Meta, SpaceX or Google at the frontier. Both outcomes result in a lower margin % at the model layer as vertically integrated model companies don’t really care where the margin $ come from. This is why it was so painful for OpenAI and Anthropic when Google was right there with them from a model competitiveness perspective and why Grok 4.5 and Muse 1.1 were just as important as Kimi K3. 
The reason Kimi K3 is only *potentially* negative for Anthropic and OpenAI is 1) the @ericvishria point that the Claude and ChatGPT products and harnesses may be more important than their models today and 2) the hypothesis that they have much more advanced model checkpoints internally that are already being used for RSI. In the latter scenario, reaching RSI even a few months ahead of other labs might be enough to cement a permanent lead. Time will tell on both points. And likely fairly quickly. Caveat would be that since Kimi K3 is not token efficient and thereby actually more expensive than ChatGPT 5.6, we may need to see a more token efficient open-source model at the frontier or see Grok 5/Composer 4/Muse 2 at multiple points on the Pareto frontier for this potential risk to Anthropic and OpenAI to play out. And I am sure they will both vertically integrate as quickly as possible while continuing the product/harness strength they have shown over the last 8 months.

👋 Sully 2.0!

You can vibe code a scribe over the weekend, but look what happens when you launch a scribe in an integrated full-suite environment for clinicians. The "AI scribe" standalone category is already dead. Writing down what happens in a visit is the easy part. Every tool does that now... The real win is an AI team that actually does the work: diagnosing, prescribing, scheduling, following up, and talking to each other. That's what we've been quietly building at Sully. One patient. One visit. A full AI team. Welcome to the future of healthcare. This is @sullyai

Every healthcare AI startup does the same thing: Build one tool. Spend 2 years chasing integration. Fail to show ROI to clients. Raise another round. Repeat. We said no. At @sullyai, we're introducing the 1st GMAH (General Medical Agent Harness) to bring the first fully functional full-suite AI agents for Healthcare. We built an entire AI team from scratch that plugs into EHRs in ONE integration and handles 30% of hospital operations on day one. 20M+ clinical tasks. 55M+ minutes back to healthcare. 96% of our clients would be very disappointed if Sully would not exist tomorrow We're not a point solution. We're the new standard.

Every healthcare AI co is building for clinicians. Nobody's building for patients. That's the $50B mistake.

@sullyai tested the major STT providers on real MENA clinical audio. 👀 Code-switching, dialect-heavy consultations, the conditions generic models fail on. 🏥 Patrick Nguyen, Head of Engineering MENA: ours was the only one that hit the performance thresholds needed for clinical documentation at regional scale.