techcrunch.com web signal

OpenAI Opens Decisions API to All, Chasing TypeSafe Jev

By Published Updated
5 sources tracking this story
OpenAI Inference Agents ai-business

TL;DR

  • OpenAI moved from DevDay preview to general availability in 23 days, a timeline the company attributed explicitly to TypeSafe Jev's commercial traction and $7.6B valuation.
  • Prompt caching at $0.01/million shrinks the 2.4x list-price gap with Jev's $0.042/million for workloads that reuse context across repeated calls.
  • Named customer benchmarks show Sierra at sub-280ms p95 latency, Datadog at 40x faster risk scoring, and Speak at a 90% reduction in time-to-action.

OpenAI opened its Decisions API to all customers on Thursday, pulling the latency floor for a single decision call from one to three seconds down to between 100 and 180 milliseconds. Rather than build a specialised classifier from scratch, product lead Nikunj Handa said the team "reconfigured" ChatGPT Luna "to speed up the inference and lower the cost."

The public release adds voice and visual modalities and geographically targeted inference resources. Early customer Speak, the language-learning app, cut time between actions from 1.4 seconds to 0.13 by swapping a language-model call for the new API.

TypeSafe's Jev defined this category and became a $7.6 billion company within weeks of launch, selling fast, low-cost classifiers that return decisions with confidence ratings. On public benchmarks, D1 Ultra from Liquid and Elephant-1 from Base Lab rank highest, with Jev at third and seventh and OpenAI's Decision Model at eighth and fourteenth. TypeSafe offerings remain the most widely queried on OpenRouter.

Handa framed the forward target plainly: a 60-millisecond call, enough for "an API call for every single frame in a video."

What others are reporting

Coverage cluster as of 24h after publish

  1. OpenAI Developer Community Read →

    Official GA announcement; confirms gpt-6-luna as the dedicated model, $0.01/million cached rate, 400-question-per-request cap, and GPT-Live voice integration.

    Build faster, more cost-efficient apps and agents that can classify text and images, evaluate inputs, and determine what happens next without generating a full model response.
  2. Unite.AI Read →

    Publishes production benchmarks from named customers: Sierra at sub-280ms p95, Datadog at 40x faster risk scoring, Speak at 90% reduction in time-to-action.

    The API returns bounded, typed answers that applications use to classify content, route requests, and prioritize work.
  3. CryptoBriefing Read →

    Focuses on the 23-day preview-to-GA timeline and enterprise compliance: HIPAA eligibility and Zero Data Retention flagged as the primary enterprise differentiators.

    The endpoint runs on OpenAI's GPT-6 Luna model. It targets routing requests, classifying inputs, and repeated micro-decisions.
  4. Firecrawl Read →

    Technical architecture breakdown placing Jev P50 at 0.21s from OpenRouter production data against OpenAI's 150ms claim, plus a staged migration framework for preview API risk.

    Constrained LLM versus purpose-built model shows up in pricing, whether output can be off-schema, and if the model can write prose.

Shared on Bluesky by 1 AI expert

Originally reported by techcrunch.com

Read the original article →

Original headline: OpenAI Opens Decisions API to All at $0.10 per Million Tokens, Chases TypeSafe Jev