Breaking

Former ChatGPT Creator Launches Jev, a Fast Decision Model 200x Cheaper Than Frontier LLMs

October 1, 2026 at 19:23 EDT

Foundation Models
AI Agents
Software Dev & Coding

TypeSafe AI, founded by ex-OpenAI researcher Diogo Almeida, released Jev, a radical departure from generative language models—a "System One" decision model that returns structured probabilities instead of text, claiming 20–200x speed and 40–400x cost advantages over frontier models. The September 2026 launch follows two years in stealth and a $40 million seed round that values the startup at approximately $200 million.

October 2026 · AI Models

Jev: Decision Speed at 1/400th the Cost

TypeSafe AI's new decision model runs 20–200× faster and 40–400× cheaper than frontier LLMs, optimized for classification, routing, and cost control.

20–200×

Faster than frontier models

40–400×

Cheaper per task

70–500ms

End-to-end latency

$0.042/M

Input tokens

Why Jev Is Faster

Parallel evaluation: Unlike language models that generate token-by-token, Jev answers every question in a single parallel pass. Adding questions doesn't increase latency.

No text generation: Returns structured probabilities and choices instead of writing prose. Eliminates hallucination risk and parsing overhead.

Synthetic training: TypeSafe's Reinforcement Learning for Calibrated Decisions (RLCD) optimizes for accurate probability outputs, not human-aligned text.

Real-World Benchmarks

vs. ChatGPT Luna 5.6

5–18× faster on classification

vs. Gemini

10–20× cheaper on email tagging

Early Tester Cost

~5,000 requests: ~$2

Latency (median)

7× faster than LLM alternatives

Known Limitations

Unreliable arithmetic, counting, and date comparisons. Designed for classification and scoring—not multi-step reasoning. Keep math in code.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.