After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution

late to reading this, but FUCK YES to playing with more shapes for intelligence! please play around with it more! append-only chat is not the final interface!

my favorite plot here was on consistency! we worked super hard to put all sorts of properties we feel are necessary for intelligent automation into the models this is also just the beginning ! 🤘

How to build a reliable risk agent without a frontier model ($0.02/sweep): • @youdotcom search • @typesafeai Jev for typed judgments • @QwenDevs for proposals & synthesis • MCP for integration Full architecture & cost breakdown below 👇

not you too jev!

Inspired by @EGafni’s Twitter thread on combining Jev with PageIndex. We show how to build long-document search with @typesafeai Jev + PageIndex. No vector database. No embeddings. Open source: github.com/VectifyAI/jev-doc… 🧵👇

me to our GTM guy: do enterprises want benchmarks him: no, the fast ones have already benchmarked their internal use cases. the slow ones are copying the fast ones. he's onboarded a quarter of the fortune 500 already 🥵

super excited for this - should make workflows much easier to build!

Jev is a phenomenal reranker. At 75% of the previous reranking cost, it improved our main-app search by 30.64% in matches per 200 candidates on searches recruiters actually run. Most recruiting benchmarks measure precision and accuracy across just 10 candidates. That isn’t representative of real recruiting and sourcing workflows, where you may need to source hundreds of candidates to make a hire. We’ll release the full benchmark soon. We're excited to work with @hmartenjoyer, @mathfax , and the rest of the @typesafeai team to continue to bring Jev to more Wrangle features. Stay tuned 👀

Is no one gonna talk about the fact that the surname in reverse reads "No Slop"

We ran agents on a site with and without Jev to see whether a decision model makes a site more accessible and usable for agents. Across browser-use, WebMCP and NLWeb: 3.6× faster on average (up to 4.3×), 7.7× cheaper on average (up to 13×). Faster in every one of the paired runs. Task success held. 240 runs, full method and data: ora.ai/blog/evaluating-jev

she's roasting me for taking 2 years AND leaking stuff on podcasts 🤦

who tf is leaking our product roadmap to MY MOM

begun, the clone war has jk, I love openai and think more competition and validation is great for developers! (assuming the model is good - plz make it good!) hopefully this is a sign for the future that building in a system one compatible way is the future

It's been almost 2 weeks since we've launched! And due to overwhelmingly popular request, we're making the datasets released with evals.typesafe.ai easier to work with! We do this in the spirit of openness, but I am still anti-public benchmarks. In that spirit, we deprecate all datasets we evaluate on (internally or publicly). (1/3)

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.

Three thoughts after playing with Jev: - this is way more than a simplistic classifier: you can pack a lot of complex state and do more than you’d think - it’ll take a minute to think about your problems in a non-LLM shape - not quite beating frontier LLMs in correctness most of the time so can’t quite switch yet but it’ll clearly get there - this is definitely a new model type we will all use by next year; I wouldn’t ignore it