ChatGPT Pioneer Releases Jev Model for Programmatic Logic

TypeSafe, founded by a co-inventor of ChatGPT, has left stealth and is launching its Jev model to automate programmatic decisions with a parallel sampling architecture.

Software systems requiring automated deterministic logic can now bypass conversational language models following TypeSafe AI’s release of Jev, a specialized System One model designed to execute structured probabilistic decisions directly within production codebases.

Developed stealthily over two years by OpenAI veteran and ChatGPT co-inventor Diogo Almeida, the model abandons text and string generation entirely.

Instead of generating sequential text tokens in an autoregressive manner, the system takes an unstructured state as input and generates type-safe structured values ​​in a single parallel query. The architecture aims at deterministic code integration, avoiding syntactic failures and output hallucinations by design.

Diogo Almeida, founder of TypeSafe, said: “Think of Jev as a frontier intelligence function: unstructured state input, written probabilistic decision output.”

Jev Model Architecture: How Hardware-Aware Parallel Sampling Works

Engineers built the platform around an alternative training methodology called reinforcement learning for calibrated decisions (RLCD).

Conventional frontier models are trained using reinforcement learning with human feedback (RLHF) or reinforcement learning with verifiable rewards (RLVR) to satisfy human conversational preferences or isolated programmatic controls. TypeSafe trains Jev to produce calibrated probabilities in the execution logic, ensuring that the confidence scores returned directly correlate with the accuracy of the output.

Starting from autoregressive token generation, Jev relies on a hardware-aware parallel sampler that evaluates and delivers all structured values ​​simultaneously.

Structural constraint restricts results to predefined schemas, eliminating the analysis pipelines and barrier wrappers required when deploying traditional large language models in business automation flows. The internal sampler handles high cardinality selections of up to 255 discrete options, using a two-stage process that scores options independently before selection.

Internal evaluations conducted from the team’s West Coast servers record end-to-end response latencies between 70 and 500 milliseconds, compared to 3 to 329 seconds seen in conversational frontier deployments.

Published workflow evaluations performed via fixed compute graphs (measuring multi-step decision branching against GPT-6 Astra and Fable 5.1 consensus baselines) recorded execution speeds up to 193.6 times faster.

Input processing costs are priced at $0.042 per million tokens, below standard conversational fees of $0.20 to $10 per million tokens. Because parallel sampling generates structured states without autoregressive token generation passes, TypeSafe provides output tokens without measurement charges.

From Doom Bots to Petabyte Data: Testing Jev’s Real-World Deployment

Production demos show Jev solving branching rules in high-speed game states and web traversal trees. In dynamic stress tests, engineers operated a real-time reactive robot on the state structures of the Doom game at 10 queries per second, generating operating costs of approximately $7 per hour.

A secondary test of Wikiracing evaluated link selection through dense encyclopedia directories, finishing tours in fewer steps than non-reasoning external models by evaluating options without generating mind-bending dead ends.

Field testing confirms core target applications through real-time feature extraction, petabyte-scale data workflows, output verification layers, and automated branching logic where handwritten rules are fragile.

TypeSafe today opened early access for developers and began recruiting engineering teams from its deployment waitlist.

See also: Microsoft AI opens review on AI Humanistic Code of Conduct

Want to learn more about AI and big data from industry leaders? Check out the AI ​​& Big Data Expo taking place in Amsterdam, California and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, including Cyber ​​Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.