[CLSTR]
< Back to situation

[REVISION HISTORY]

TypeSafe AI launch of Jev model and Laya response

Updated 3 times since CLSTR started tracking revisions of this situation.

What changed

2026-09-21 16:03 UTC → 2026-09-22 11:20 UTC · added removed

TypeSafe AI, a San Francisco-based startup led by former OpenAI researcher Diogo Almeida, has launched Jev, a System One Model ‘System One’ model designed for software automation and machine interaction. Unlike traditional Large Language Models that focus on text generation, Jev provides structured, typed probabilistic decisions based on predefined schemas using three primitives: choice (selecting from a set), score (positioning on a rubric), and noul (a calibrated yes/no). The model utilizes Reinforcement Learning for Calibrated Decisions (RLCD) to optimize for epistemically honest probabilities. This architecture allows for parallel sampling with response times between 70ms and 500ms. By prioritizing calibrated confidence scores over generative reasoning, the model aims to eliminate traditional hallucinations and type errors. TypeSafe AI claims Jev is significantly faster and cheaper than frontier LLMs, citing a cost of $0.042 per million input tokens with free output tokens. The company has raised $40 million to support this machine-native intelligence layer. Following the launch, Nandakishor Mukkunnoth, founder of ConvAI Innovations, released Laya, an open-source Apache-2.0 response to Jev. Mukkunnoth alleges that the non-autoregressive decision concept used by TypeSafe was previously detailed in his March 2025 research, though he characterizes the dispute as “uncredited prior art rather than copied code.” Laya includes both English and multilingual checkpoints. Technical implementations, such as through the Vercel AI Gateway using FastAPI, Gateway, demonstrate Jev’s utility in triaging support tickets by determining urgency, department, and priority levels. While early tickets. Early testing suggests Jev is 5 to 20 times more efficient than traditional models for specific tasks, though developers note that applications must still maintain control over contracts, policies, and fallback mechanisms.

Versions

  1. 2026-09-22 11:20 UTC TypeSafe AI launch of Jev model and Laya response
  2. 2026-09-21 16:03 UTC TypeSafe AI launch of Jev model and Laya response
  3. 2026-09-20 09:30 UTC TypeSafe AI launch of Jev model and Laya response
  4. 2026-09-18 23:27 UTC TypeSafe AI launch of Jev model

Only revisions since CLSTR began indexing content versions appear here. Select a version to see what changed compared to the one before it.