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
we announced Jev 1.14 on 10.14 at 1:14pm PT 🔥
a new and improved model, along with experimental features for our dreamers:
• images
• a new box type
• longer context
surprise, Jev was already able to see you! 👁️
BUT our bar for production-grade intelligence is quite high, so builders beware, these are preview features.
enter the ~*~ danger zone ~*~ at your own risk. 🚨
Jev can serve as better natural language search on websites.
It can scan thousands of Zillow listings and classify properties by things you can't normally filter for - e.g. architecture, renovation status, proximity to freeways.
This was done in <20 sec and costs $0.18 👇
Breaking: Browser Use + Jev = Ultrafast ⚡
Findings flights took 7s and cost only $0.0039 🤯
> new action space every step
> DOM state space
> small LLM fallback to type
(this video is at 1x speed btw)
Built a tiny open source browser agent. try it below ↓
how to add a decision model to your AI workflow
1,000 yes/no decisions can cost around:
~$10 on GPT-6 Astra
~$4 on Claude Opus 5.5
~$0.04 on JEV
use language models for planning, writing and coding, and decision models for decisions
a decision model can:
> check your agent’s output before it ships
> approve a draft or request revisions
> route a task to the right agent or tool
> classify leads, tickets or comments
> score fit, quality or priority
> check whether content follows your brand rules
> pick an option from a shortlist
> decide whether to continue, stop or request a review
at a much faster speed as well than any regular LLM
here’s how to add one
1. find the decisions in your workflow
look for repeated steps that need a choice, classification or score
start with one decision you can compare against examples you’ve already reviewed
2. define the question and allowed answers
give the model the information and references it needs
for a content workflow, that could include your draft, voice DNA file, anti slop rules and approved examples
ask it to return a choice such as approve, revise or human review
3. connect the answer to an action
your software uses that answer to decide what happens next
approved drafts move to your review queue, revision requests return to the writing agent, and uncertain cases go to you
4. test before letting it act
compare its answers with your reviewed examples
if it returns confidence scores, check whether those scores match how often it gets the decision right
start with actions you can easily reverse
5. put simple checks first
use code for fixed rules, such as required fields and character limits
then use the decision model for checks that need context, before starting more expensive work
6. track quality and cost
measure cost per completed task, incorrect decisions and how often another model or a person needs to review
use the failures to improve your questions, references and allowed answers
I built a browser extension with Jev @typesafeai that can hide/collapse posts on X based on natural language.
It's so fast that it's not noticeable and insanely cheap...this must be the future of "ad blockers" and content firewalls.
jev is insane 🫣
it makes realtime virtual try-on hauls possible.
built this experiment for Drape with @typesafeai
> i talk
> jev reads transcript + what i'm wearing
> picks from my closet
> changes my outfit in realtime
cost: $0.0011 per decision
time: ~620ms per decision
imagine getting ready like this:
Jev + Astra beats the Ender Dragon in Minecraft in 8 minutes 43 seconds! ⏱️
Cost less than $1 ($0.01 Jev, $0.96 Astra)
I open sourced the code and explain the harness setup below. This type of movement is only possible with Jev's near instant decisionmaking, and some continually learning skills from Astra.
Jev + on-device models = results in seconds with no LLM in the loop.
Quick demo app to show the possibilities. Drop in an audio file: Ear detects the language, Voz transcribes it and Redact removes PII.
Then @typesafeai's Jev makes about 20 decisions in one call in milliseconds, and picks which of our on-device models to run. Voice memo to to-do list. Meeting to redacted transcript. Podcast to clips.
i’m probably going to regret this.
10,000+ videos were imported by our users into Ghostfeed in last 6 months.
this weekend, we picked 1,066 reaction videos and analysed all of them with JEV + GEMINI.
now you can search by structure, reaction, and who’s on camera.
opening the library today which has been our months of data gathering, clone all videos to mass publish them on instagram, tiktok and youtube.
comment "reactions" if you want the access to the library.
I built a Sales Copilot using Jev
→ Listens to sales call live
→ Tells you what to say next
→ Helps you follow the script and handle objections
→ Shows you what stage of the call you're in
→ Gives you live signals and probability of closing
Demo below on a recorded sales call:
jev is INSANE.
in 40 seconds it broke down 724 live ads from 37 brands.
every hook. every format. offer. cta. awareness stage. landing page mismatch. used 9 cents of tokens.
(will be avail in @stealads + mcp)
got @typesafeai’s new model Jev to play Super Mario Bros.
fast inference + structured outputs makes it surprisingly good for real time use cases.
I'm excited to see what can be done with these new models!
Jev + SuperX = virality solved ✅
Every post gets 61 questions in ~1s for $0.0004 🤯
> fitted on 9,481 real posts from 207 creators
> picks the viral post 2 in 3 times
> never rewards reply bait
So: write, score, rewrite, stop when it peaks.
Free, no signup. try it below ↓
1/ Jev, a decision model by @typesafeai, sparked a burst of projects and discussion. We tested it using Ori Eval against popular LLMs on OpenRouter at judging.
Jev was >5x faster than the next fastest model, and even its slowest requests beat every other model's median.
Jev + LinkedIn bookmarks = your saved posts are finally searchable ✅
479 posts classified in 225 seconds for $0.0195 🤯
LinkedIn has no native way to search what you saved Jev reads every post and tags the topic, hook, and format you can finally find that post you saved 6 months ago
So: save, classify, search. Never lose a good post again.
Free and open source. try it below ↓
We built a plugin that gives Jev a browser in Cline, and have been blown away by the results.
1. Install it in our new desktop app: Customize > Marketplace > Plugins > search 'jev-browser'
2. Create a Vercel AI Gateway API key, then save it to ~/.cline/plugins/cline-jev-browser.config.json as {"gateway": {"apiKey": "..."}} and restart Cline.
3. Ask any browser task and it will launch Chrome in the background to complete it.
🎇 I made a small Chrome extension: a side panel that drives any site's WebMCP tools with Jev!
When you type, on every keystroke it picks the relevant page's tool, fills in the arguments, and tells you how sure it is.
Here it is grocery shopping:
JEV is INSANE.
We gave it 700 high-intent leads and personalised outreach messages.
In 40 seconds, it predicted how each message would perform, assigned a confidence score and detected lead-message mismatches.
All for just $0.09.
JEV can also score leads, analyse buying signals, match each prospect with the best message and identify the campaigns most likely to perform based on data.
Coming soon to @GojiberryAI+ MCP.
Comment “JEV” for early access.
I gave Jev 3,282 of my X posts across 100M views and asked it to find what actually works for growth.
4,252,330 tokens $0.1282 for the full 8m 34s run!
Each post got 8 questions about the topic, hook, tone, whether it teaches something, etc.
How-to posts got 150 median likes vs the average median of 44.
AI and coding was a 1.9x multiplier topic compared and SEO, despite recent posts, was right at base median 1.0x - surprisingly.
The recommended topic + angle + voice formula was:
AI coding + teach something + provocative
We just ran Jev on our WebMCP benchmark.
The result: basically broke the benchmark.
Jev + Mercury 2.5 (a fast, low-cost LLM) using WebMCP solved 100% of the tasks at roughly 112× lower model cost than GPT-6 Astra using computer use with code execution. Compared to Astra using screenshot-based computer use, the model cost was 245× lower (!).
We also compared Jev operating the browser with and without WebMCP.
We used Browser Use’s open-source Ultrafast, with some improvements to the harness to make it more reliable across the benchmark.
Jev’s browser-control accuracy on its own was not amazing - adding WebMCP nearly doubled the number of solved tasks, from 25/49 to 49/49, while reducing model cost by 18% (more on why below).
The benchmark and methodology are fully open and reproducible.
Full results: https://webmcp.com/benchmark
A few words on how the Jev + WebMCP harness works and why this is exciting:
Jev receives text as input and a set of discrete options it can choose from. With WebMCP, those options are the tools exposed by the website. At each step, Jev sees the task, the available tools and previous results, then picks what to do next.
The limitation is that Jev can’t generate arbitrary text, which you need for tool arguments. For example, it can choose the search_products tool, but it can’t generate the search query itself.
So we split the work: Jev picks the tool and Mercury 2.5 generates the arguments if needed.
This works well because turns out most of the cognitive load in these tasks is around choosing the right action. The argument generation itself is relatively simple, so we can delegate to a small and very fast model. We used Mercury, which outputs 1,000+ tokens/sec and is very cheap.
The result is a pretty simple combination: Jev for tool selection + Mercury for arguments + WebMCP for the interface. It ends up being very reliable, very fast, and very cheap.
A few words about Ultrafast and why do we think it underperforms:
Without WebMCP, Jev chooses from the page’s controls: which button to click, which field to fill, or which option to select.
But choosing a valid button is different from choosing the right next step. The agent still has to navigate menus, understand forms, recover from errors and recognize when the task is actually complete.
Our hypothesis is that WebMCP makes the decision space much simpler. Instead of figuring out a sequence of clicks through a website, Jev chooses explicit actions that directly advance the task.
@typesafeai itself documents weaker accuracy on questions requiring multiple reasoning steps. WebMCP moves much of that complexity into the website’s tools, leaving Jev with clearer decisions and fewer opportunities to go wrong (in a sense WebMCP "compresses" a sequence of clicks into one tool call).
Our modified Ultrafast setup solved 25/49 tasks - that is a result for our particular implementation and benchmark, not a universal limit on Jev or Browser Use. We are open to more harness optimization to get this result to perform better, feel free to directly contribute to the benchmark here: https://github.com/nekuda-ai/WindTunnel
Browser-use ultrafast: https://github.com/browser-use/jev-ultrafast
Just created this with Jev by @typesafeai. A live viral post analyzer. As soon as you stop typing for .5 seconds it analyzes the viral potential.
Going to try and actually make this good, will need to scrape a lot of twitter data...
Notice how it also categorizes the tweet live... I could have it surface similar tweets on the right side for inspiration... idk just experimenting.
Types Beyond Decisions: box
We’re also releasing a new type that is somewhere in the gray area of decisions: a single bounding box. We warned you there’s a reason we call them System One Models!
(only on jev-preview)
Diogo Almeida of TypeSafe on the new Box type: one bounding box per question, somewhere in the grey area of decisions, and only on jev-preview.
Kyle Jeong
@kylejeong
we built blazing fast computer/browser use with Jev + @Stagehanddev.
this task cost $0.001 and executed at near instant speed (in a remote browser btw)
the loop: observe the page, send a11y tree as state + actions as questions, Jev decides the next action, then Stagehand executes it.
What people posted on X while they built with Jev. Every card plays the original video or shows its images here, so you see the demo before you open the thread.
