Jev and the Arrival of Model Tetris: Decision Models Get a Standard API

On September 15, startup TypeSafe AI shipped Jev, a model that generates no text at all. Within three weeks, more than a dozen imitators followed from OpenAI, Cloudflare, Perplexity and open-source projects, and on October 27 Liquid AI added a class of models small enough to run on a phone. The industry is no longer playing LLM Tetris. It is playing model Tetris.

Jev is what TypeSafe calls a System One model, a term borrowed from Daniel Kahneman's Thinking, Fast and Slow. Founder Diogo Almeida, who says he helped build the instruction-following research at OpenAI that became ChatGPT, spent two years in stealth on one question: if models have been superhuman at chat for years, where is all the automation? His answer targets the fast, intuitive decisions software makes all day — route this ticket, score this risk, take this branch.

What does a System One model actually do?

You hand it a piece of state and typed questions; it returns structured answers with calibrated probabilities. There are three question types: a Choice among options you define, a Score against a rubric you write, and a yes-or-no probability TypeSafe calls a Noul. The launch pricing is aggressive: $0.042 per million input tokens, output free, responses in 70 to 500 milliseconds.

One claim needs sharpening. TypeSafe says Jev cannot hallucinate, meaning it cannot return an answer outside your schema. It can still pick the wrong answer from inside it, and it will report how confident it was when it did. The name itself is a nod to William Stanley Jevons: every drop in the cost of intelligence unlocks more uses for it.

How did one product become a category in three weeks?

The vendor sequence came fast:

• At DevDay in late September, OpenAI previewed a Decisions API built on GPT-6 Luna; on October 6 it opened it to all developers at $0.10 per million input tokens, claiming decisions roughly ten times faster than the same model through its regular API.
• On October 1, Cloudflare released Clef, open-weight decision models whose smaller version returns a median decision in under 40 milliseconds.
• The same day, Perplexity opened its own Decisions API on open weights, claiming a narrow win over Jev on its own benchmark panel.
• Amazon's Strands Labs reportedly shipped an open-source decider, and Liquid AI's d1 and Upstage's Solar Decide joined the open-source list.

On October 2, llama.cpp added an endpoint speaking the request format Jev introduced; Cloudflare's Clef speaks it too. The interface standardized before the market did — the same dynamic that made container runtimes interchangeable. Published prices range from $0.02 to $0.24 per million input tokens across vendors, and TypeSafe itself says it cannot prove its pricing isn't subsidized.

What did a hands-on test show?

The author wired MAME, the arcade emulator, into an agent harness via Claude Code, then built three Pac-Man planners on one tactical loop: code-planner (plain code), jev-planner and haiku-planner (Claude Haiku). Each ran 100 times. The game sends a new board state roughly every 17 milliseconds, so latency mattered.

Planner	Mean score	Best score	Cleared level 1	Median frames missed
code-planner	5,100	15,640	51%	0
jev-planner	4,499	9,480	27%	21
haiku-planner	3,406	9,090	10%	59

The outcomes rank exactly with latency. Jev likely beat Haiku because it answered sooner, not because it was smarter. A correct answer delivered after the window closes is a wrong answer — true for card authorizations, ad bids and agent-loop tool dispatches, even without a flashing GAME OVER.

How should CIOs sort a decision?

The ordering: known, stable rules mean write code; a known option set with fuzzy rules means a System One model; written or explainable output means an LLM and the wait. Four questions sort any decision: output shape, latency budget, explainability (a System One model returns a probability, not a rationale) and stakes.

The caveats are real. Nearly every benchmark in the category is vendor-reported, and TypeSafe's own workflow evaluations were written by its own team. A probability with no rationale is an audit problem in regulated work, so expect these models to route and screen rather than decide alone. And where no deadline exists — contract review, quarterly forecasts — the better answer should win. Fast doesn't beat smart. Fast beats late.

The strategic point: keep model choice behind an interface you control. Databases have had 60 years; agents have had 18 months; System One models are two months old. Lock-in used to be denominated in egress fees. Now it is denominated in model usage patterns.