Jev: From Launch to Viral Success in Three Weeks

Episode 1 · TypeSafe's Jev: A Text-Free Decision Model That's Fast and Dirt Cheap (2026-09-28, 6 posts)

TypeSafe AI opened early access on Sept 15 to Jev, a "decision model" that sparked heated discussion on Hacker News. Created by founder Diogo Almeida, an OpenAI veteran and InstructGPT co-author backed by a $40M seed round, Jev differs fundamentally from mainstream LLMs: it does no text generation and accepts only three question types—Choice (up to 255 options), Score (2–10 bands), and true/false—returning decisions directly.

Confirmed

• Input pricing as low as $0.04 per million tokens, with claimed 190x faster and 440x cheaper performance versus frontier LLMs (some posts cite 193x/444x)
• @every reported a real use case: Jev-based news classification expanded morning screening from 20 to 500 items
• @sanjaykalra called it the most useful model release of the quarter for CIOs—provided it is used for the narrow tasks it was designed for—and noted Jev uses a cascading architecture
• @大模型之路 explained its popularity in the Agent community: most Agent requests in call logs are judgments (classification, gating, scoring) rather than content generation, yet they pay full LLM latency and token costs; Jev peels these small judgments off the LLM bill
• @大模型之路 also cautioned that the 193x/444x figures are marketing numbers that deserve skepticism

Why it matters

Jev represents a "System 1" narrow-domain model route: for massive, low-value subjective judgment tasks (is this email urgent, does this paragraph look AI-written), a specialized model can replace expensive LLM calls and scale cost-sensitive pipelines by an order of magnitude. @every's news-screening case and @大模型之路's cost analysis both point to this scale effect. Authors also stress its limits: scenarios beyond the three question types, or requiring generation or complex reasoning, are not a fit—and the advertised multiples should be taken with a grain of salt.

• Jev, a classification-only model from TypeSafe, scales a news workflow from 20 to 500+ items — every · 2026-09-28
• TypeSafe's Jev Decision Model Claims 190x Speed, 440x Cost Savings vs Frontier LLMs — sanjaykalra · 2026-09-28
• TypeSafe AI's narrow model Jev answers just three question types via cascade architecture — sanjaykalra · 2026-09-28
• JEV, the 'decision model' that writes nothing, ships — but 193x speed claims need discounting — 大模型之路 · 2026-09-28
• Jev, a Word-Free System 1 Model, Outputs Only Decisions at 4¢/M Input Tokens — Matt Wolfe · 2026-09-29
• JEV decision model:剥离Agent高频小判断，省下大模型延迟与token费 — 大模型之路 · 2026-09-29

Episode 2 · TypeSafe's Fast Decision Model Jev Goes Viral as OpenAI Chases (2026-10-08, 7 posts)

TypeSafe AI (co-founded by Erik Gafni, Sasha Sheng, and Diogo Almeida) released its fast decision model Jev in September. Within three weeks of launch, it was processing 1 trillion tokens per day, and roughly a quarter of the Fortune 500 have already integrated it. According to Fortune, the model has quickly gained popularity among AI developers and in Silicon Valley circles, and OpenAI is pursuing a similar direction.

Confirmed

• Jev is TypeSafe's first "System One" model, inspired by Kahneman's Thinking, Fast and Slow: System 2 is slow and deliberate, while System 1 is fast and automatic; most of today's agentic systems lack this fast pathway (@veskost)
• Jev is not a traditional LLM—it doesn't generate text, but instead outputs structured judgments with probabilities, including Noul judgment statements, Choice options, and Score ratings (@大模型之路)
• Official and reshared claims: 200x faster and 400x cheaper on classification tasks, able to decide an agent's next action in milliseconds at near-zero cost (@LangChain, @Arindam1729)
• LangChain published a blog post introducing Jev and explaining how to embed it into agent loops

Not Yet Confirmed

• Performance claims such as "200x faster, 400x cheaper" come from TypeSafe and have not been independently verified
• The specific progress of OpenAI's reported "close pursuit" mentioned by Fortune remains unclear

Why It Matters

• @veskost points out that in the era of large models, developers have forgotten that classifiers can be fast, cheap, and useful; the pain points of traditional classifiers lie in training, monitoring, updating, and system syncing, and letting agents build and maintain Jev-style classifiers themselves can replace fine-tuning—a pragmatic engineering path
• @Arindam1729 cautions that most developers will use it in the wrong scenarios, and offers 10 project use cases suited for Jev, showing that knowing its applicability boundaries is equally critical
• If millisecond-latency, near-zero-cost decision models see wide adoption, they could upend the default architecture of calling a large model at every step in agent loops, prompting giants like OpenAI to follow suit

• Decision model JEV hits 1T tokens daily, quarter of Fortune 500 on board — 大模型之路 · 2026-10-08
• Jev-Style Classifiers: Letting Agents Build Fast, Cheap Classifiers Instead of Fine-Tuning — vesko_st · 2026-10-09
• TypeSafe's Jev: Agents Building Fast, Cheap Classifiers as Their System One — vesko_st · 2026-10-09
• TypeSafe AI's Jev decision model claims 200x faster, 400x cheaper classification in agent loops — LangChain · 2026-10-09
• Jev pitched as the fastest AI model for agents: millisecond decisions at near-zero cost — Arindam_1729 · 2026-10-09
• Jev, a fast-decision AI from TypeSafe AI, goes viral in Silicon Valley as OpenAI follows — jeremyakahn · 2026-10-09
• TWIML podcast: TypeSafe's Jev model bets on machine-native intelligence over LLMs — samcharrington · 2026-10-10