A Jev-as-a-Judge paper reports that accepting TypeSafe AI’s confident JEV verdicts and escalating the rest to GPT-6 Astra kept 99% of GPT-6 accuracy on 510 held-out preference pairs, at about 57% of GPT-6’s fee. The authors say the escalation threshold did not transfer to every fallback model and should be set on one’s own data.
Pydantic says the agents use TypeSafe AI’s Jev classifier to answer typed questions, returning each agent’s existing output model with one confidence score per field.
On September 22, 2026, Ahmad Awais said he was launching Jev on Command Code for free, limited to headless mode or the Provider API. He also open-sourced the mod cmd-mod-jev-nudge.
Jev can be used only in headless mode or through the Provider API. Ahmad Awais also open-sourced the mod cmd-mod-jev-nudge and says the free access is for keeping an agent running while more work remains.
Shahriar Tajbakhsh said Metaview added TypeSafe AI's jev to every agent over the weekend. He said candidate searches in its sourcing product now take seconds instead of minutes, with the same accuracy, about 10 times the speed, and a meaningfully lower cost per search.
LLM requests can be classified automatically with typesafeai's Jev in OpenRouter Classifiers, and the results can be analyzed in Explore.
TypeSafe quoted MotherDuck’s claim that prompt_jev(), a SQL function powered by TypeSafe’s Jev model, made text classification about 50x faster at about 1% of the cost. MotherDuck reported 100k rows in 40 seconds for $0.50 versus 32 minutes and $37 for an LLM.
TypeSafe AI said Jev is now available and linked to https://console.typesafe.ai. The post did not specify what Jev is or any access terms.
Nathan Drezner says TypeSafe AI's Jev middleware can be used with LangChain managed deep agents for model routing. Users define routes, and Jev handles model selection.
LangChain said it tested Jev against GPT-5.6 Luna, GPT-5.6 Terra, and Claude Sonnet 4.6 as agent-eval judges, reporting a human-reviewer match on every call, up to 913x lower variance, 0.44s per call, and a $0.34 full-run cost versus $28.17 with Claude Sonnet 4.6.
LangChain highlighted a sub-10-minute video by Sydney Runkle breaking down @typesafeai’s new Jev model and how to plug it into agents.
Elvis (@omarsar0) said he used Jev to organize about 2.3K AI research papers for a total cost of $0.14 in about 83 seconds. He said Jev agreed with 75% of prior DeepSeek V4 Flash tags, found about 579 high-confidence topic changes, and that a 30-item manual check of disagreements was fully accepted before the changes were applied in production.
TypeSafe AI said Jev is now available to everyone and has no waitlist. The company also said all users start with $5 in credit (~120 million tokens).
In a quick demo, TypeSafe AI's Jev scored 80.5% exact at about 200 milliseconds and $122 per 1M, against Luna's 84.9% exact at about 2 seconds and $482 per 1M.
