OpenAI now sells a Jev-shaped endpoint to everyone
OpenAI is now selling the same kind of typed decision call that TypeSafe's Jev sells, and it adds image input and European data residency. OpenAI Developers said on X on 6 October US time (7 October in Singapore) that the Decisions API is in public beta for all developers.
The Decisions guide says the API returns typed answers "about 10x faster than the Responses API". The X post words it as "up to 10x faster than GPT-6 Luna through the Responses API". Neither gives a latency in milliseconds. OpenAI says it expects general availability "in the coming weeks".
At DevDay on 29 September the API was a limited preview with no price. See OpenAI DevDay 2026 brings Ultrafast, a Decisions API preview, Codex cloud and plugin extensions.
Jev Ultrafast runs on TypeSafe alone today
Jev matters here because Browser Use's open browser agent uses it to pick every click. See Jev Ultrafast: Browser Use wires TypeSafe Jev into an indexed browser-action loop.
When The Frontier read the Jev Ultrafast README at 19:46 SGT on 7 October, setup still asked for a TYPESAFE_API_KEY and an OpenRouter text key. The README names no OpenAI backend for the decision step.
The README also says the default loop sends "no screenshots" and Jev "consumes structured state". OpenAI's image input therefore does not touch that loop as written.
The Frontier found no post about the Decisions API on the X accounts of TypeSafe or Browser Use at 19:47 SGT. Neither company's plans in response are known.
Price, inputs and limits side by side
Both products answer the same three question types. OpenAI calls them predicate, choice and score. TypeSafe calls them Noul, Choice and Score. The table compares OpenAI's guide and pricing page with TypeSafe's models page, read at 19:46 SGT.
| Item | OpenAI Decisions API | TypeSafe Jev 1.13 |
|---|---|---|
| Model | gpt-6-luna only | jev-1.13.0 |
| Input price per million tokens | $0.10 | $0.042 |
| Output tokens | Not charged | Free |
| Input types | Text and inline base64 images | Text only |
| Published limits | None stated in the guide | 64k context, 80 requests per second |
| Data controls | Zero Data Retention and HIPAA for eligible customers, US and Europe residency | Zero Data Retention for enterprise customers |
Jev's list input price is about 2.4 times lower than the OpenAI rate. Through the Responses API, gpt-6-luna costs $0.10 input and $0.50 output at standard short-context rates. The new endpoint drops that output charge.
TypeSafe's launch post says its service is "currently based" on the West Coast. Its models page warns that rate limits "can change without notice".
No shared benchmark yet
OpenAI publishes no accuracy results, no latency figure and no rate limit for the endpoint in the guide, the changelog or the post.
TypeSafe's launch post claims responses of 70ms to 500ms. It says its workflow evals use GPT-6 Astra and Fable 5.1 as the reference, which "biases answers towards OpenAI and Anthropic's models". Both sets of numbers are self-reported.
Tessl reported on 23 September that Jev ran its verifier suite 13.6x faster and 2.7x cheaper than GPT Luna 6 as a judge. The two agreed on 85.9% of decisions. That test used Luna with written reasoning. The Decisions endpoint and Tessl's judge setup are different, so neither result predicts the other.
OpenAI's September changelog also added computer use to the Agents API in "an OpenAI-hosted browser". That puts OpenAI next to Browser Use's own hosted agents, which the README lists behind a waitlist.
Test on your own labelled data
- Jev Ultrafast users have nothing to change today. The repo still calls TypeSafe.
- Teams that need image checks or EU data residency can trial the Decisions API, since Jev takes text only.
- Teams pricing on volume can compare $0.10 with $0.042 per million input tokens against their own traffic.
- OpenAI's guide says to set thresholds from labelled examples. Run the same set through both before switching.
Perplexity sells a third hosted option at $0.04 per million input tokens. See Perplexity Decisions API and pplx-decider-v1-27b: $0.04 per million input tokens, Apache 2.0 weights.