What happened
A new class of AI models called System One Models just launched, and the first one, named Jev, answers certain questions up to 200 times faster than a regular chatbot like ChatGPT. The person behind it isn't a random startup founder — it's Diogo Almeida, one of the researchers who co-authored the RLHF technique and the original InstructGPT paper, the research that shaped how ChatGPT was trained back in 2022. In 2024 he started a new company called TypeSafe AI, spent two years building quietly out of public view, and just came out of stealth with Jev.
Here's the core idea, in plain terms. When you ask ChatGPT a question, it writes out an answer one word (technically, one "token") at a time, like someone typing a sentence live. That's great for writing an email or explaining a recipe. But a huge number of behind-the-scenes AI tasks don't need a written sentence at all — they need a single decision. Which support ticket category does this message belong to? Is this transaction risky or safe? What should an automated assistant do next? A normal chatbot still writes out a full sentence to answer these questions, and then a separate piece of software has to read that sentence and extract the actual decision from it. Jev skips all of that. It goes straight to the answer — the category, the risk score, the next step — along with a confidence percentage for each option, without ever generating a sentence.
Because it skips the word-by-word writing process, Jev responds in 70 to 500 milliseconds, which TypeSafe AI says is 20 to 200 times faster than standard large language models doing the same kind of job. Pricing is unusual too: $0.042 per million input tokens, and the output — since there's barely any text to output — is free. The model is available now in early access. You can read the original announcement at TypeSafe AI's blog, and it's already been covered by outlets like AI News and The Rundown AI.
What it means for you
This sounds technical, but the effect shows up in everyday things faster than you'd think. A few real scenarios:
At work: If your company uses a chatbot to route customer emails to the right department, a model like Jev could make that routing near-instant instead of taking a second or two per message — which matters a lot when you're processing thousands of tickets a day.
Running a small business: Say you sell online and want to flag risky orders before they ship. Instead of paying for a full AI model to "think out loud" about every transaction, a decision-only model can just tag each order as low, medium, or high risk in a fraction of a second, for a fraction of the cost.
Building an app or side project: If you're a solo developer or a student building an app that needs an AI "traffic cop" — deciding what a user's message is about before sending it to the right tool — this kind of model does that job cheaper and faster than routing everything through a full chatbot.
Creative work: If you build interactive stories, games, or chatbots for fun, these models are well suited for the constant small decisions behind the scenes (what mood is the player in, what should the NPC do next) without slowing down the experience with visible "typing."
Income and freelancing: If you offer automation services — building bots, workflows, or backend logic for clients — being able to say "I can make your system's decision-making 20 to 200 times faster and cheaper" is a concrete selling point you can bring to a client conversation today.
At home: Less directly, but still relevant — the smart assistants, spam filters, and app recommendations you already use quietly make thousands of small decisions like this every day. Models like Jev are part of the plumbing that could make those feel snappier.
How to try it right now
You don't need to be a programmer to get a feel for what's changing here, but actually using Jev itself does require some technical setup since it's built for developers, not a chat window you type into.
1. If you just want to experiment with AI without any setup or cost, start with free browser-based tools first — for quick tasks like summarizing text, generating images, or drafting content, a hub like mykreatool.com gives you several free AI tools in one place with no signup friction, which is a good way to get a feel for what current AI can do before diving into anything technical.
2. If you're a developer or work with one, go to TypeSafe AI and request early access to Jev. It's positioned as an API you plug into your own app or workflow — the same way you'd connect to any AI service — not something with its own consumer app.
3. Read the technical announcement first. The official blog post explains the input/output format Jev expects, which matters if you or your developer are wiring it into an existing product.
4. Start with one narrow decision task — ticket routing, content moderation flags, risk scoring — rather than replacing your whole AI stack at once. That's the kind of job this model is built for.
Upsides and what changes
The biggest shift is cost and speed for a specific, common category of AI work: classification and decision-making. Businesses running high volumes of these tasks — support routing, content moderation, fraud checks, recommendation logic — could see meaningfully lower AI bills and faster response times, since they're no longer paying for and waiting on full sentences they were only going to throw away after parsing. It also lowers the barrier for smaller teams: a $0.042-per-million-token price point with free output is cheap enough that a solo developer or small business can add AI decision-making to a product without it becoming a real line item. For the industry, it signals that not every AI problem needs a bigger, more general chatbot — sometimes a smaller, purpose-built model beats a giant one at its own narrow job.
Limitations
Jev isn't a replacement for ChatGPT or similar assistants — it doesn't write essays, hold a conversation, or explain its reasoning in a paragraph, because that's not what it's designed to do. It's an early-access product from a very new company, so pricing, availability, and performance claims come from TypeSafe AI itself and haven't yet been stress-tested across years of real-world use the way established models have. It's also a developer-facing tool, not something an average person can open and use directly, so most people will experience its benefits indirectly, through apps and services built on top of it rather than by using it themselves.
Conclusion
System One Models like Jev point to a simple but underrated idea: a lot of AI work is just decision-making, not writing, and treating it that way makes it dramatically faster and cheaper. If you build or run anything with AI in the loop — a bot, an automation, a small business tool — today's action is easy: pick one repetitive decision your system currently makes by generating and parsing text, and check whether a purpose-built model could do it in under 500 milliseconds instead. And if you just want to see what free AI tools can already do for you right now, start with mykreatool.com before committing to anything technical.
Comments 0