Azure OpenAI Decisions API hits general availability in Microsoft Foundry

Microsoft announced general availability of the Azure OpenAI Decisions API inside Microsoft Foundry. The GA arrived one day after OpenAI introduced the API on its own platform. Instead of asking a model to write "I'll choose option B" and parsing the reply, developers send multimodal or structured inputs. The API returns a discrete choice, along with probabilities and confidence scores, that code can branch on directly.

The technical pitch is determinism and speed. Agent frameworks today often spend tokens and latency generating free text just to pick the next tool or route a ticket. Brittle output parsing then adds failure modes. A scoring endpoint removes generation overhead and gives calibrated confidence, so a workflow can set thresholds. For example, it can auto-approve above 0.9 and escalate to a human below. For enterprises, Foundry GA means the API comes with Azure's compliance, networking and regional controls.

Competitively, this extends a trend toward structured outputs and function calling, but goes further by stripping generation out entirely. Perplexity's CEO this week also promoted a locally runnable "Perplexity Decision Model," suggesting "decision" endpoints are becoming their own product category. Microsoft paired the GA with expanded partner capacity credits for Copilot Studio, reflecting pressure to show Copilot monetization ahead of earnings.

Watch how well-calibrated the confidence scores are in practice, and how pricing compares with cheap small-model classifiers. Teams already fine-tuning small routers may find the API's advantage is convenience rather than capability.