OpenAI announced a Decisions API that uses its Luna model to return structured decision outputs instead of chat-style text
Product launch Updated 66% confidence first seen
OpenAI announced a Decisions API built on its Luna model, intended to provide decision-model outputs rather than prose responses. The announcement describes low-latency structured results and says the API is starting with a limited preview before a broader rollout.
Jev
OpenAI
OpenAI Decisions API
TypeSafe AI
CNNs
Calibration
GPT-6 Luna
Jev
Luna
MarkTechPost
RNNs
The New Stack
Transformers
TypeSafe
bag-of-words
Decision brief
- What changed
- OpenAI announced a Decisions API built on its Luna model that returns structured decision outputs rather than chat-style text. According to the announcement as reported by The New Stack, the API is entering limited preview first, with broader rollout planned in the coming days.
- Why it matters
- This gives teams building agents and workflow automation a new OpenAI interface optimized for fast, schema-like decision calls instead of prose generation, which could change how they design routing, gating, reranking, and verification steps. For leaders already standardizing on OpenAI, it may offer a simpler way to separate decision logic from full LLM responses; for others, it signals that decision-specific inference is becoming a competitive product category rather than a niche pattern.
- Evidence
- The New Stack directly reported OpenAI’s Tuesday announcement, including the core product change, the Luna model basis, the 150 ms reported latency, and the limited-preview launch plan. MarkTechPost and Ahead of AI independently support the broader market context by describing strong attention around TypeSafe AI’s Jev and the emerging value of decision-oriented models for agent loops and classification-style tasks.
- What remains uncertain
- The coverage does not provide OpenAI’s pricing, API constraints, reliability metrics, schema guarantees, or customer availability details, so the operational and economic impact is still unverified. It is also unclear from the reporting whether the 150 ms figure reflects typical production performance or a specific benchmark condition, and how broadly the capability will differ from existing structured-output patterns in OpenAI’s standard models.
- Monitor next
- Watch for OpenAI’s broader rollout details, especially public documentation on pricing, rate limits, supported output schemas, and customer examples using the Decisions API in production.
Analytical support, not advice — assumptions and open questions stated above.