Decisions API: Low-Latency Model for Routing, Classification and Actions
Melvin Vivas · X video post · 2026-10-22 · 1:11 · 355 views · Open on X
Topics: LLM Fundamentals, LLMOps, Deployment & Monitoring, AI Agents, Tool Use & MCP · Level: intermediate
Summary
Melvin Vivas shares a launch video for the Decisions API, now in public beta. It's a fast "decision model" that lets an app choose the right model, tool or action in near real time. The demo shows three uses: sorting loose user input into sales leads, steering a car in a game from video frames, and picking facial expressions for an animated character while GPT Live 1 handles the voice. Melvin says he hasn't tried it himself. He just notes that decision models are becoming popular.
Key points
- The Decisions API is in public beta for all developers. Its job is to let an app choose the right model, tool or action in near real time.
- It is claimed to make decisions up to 10x faster than GPT-6 Luna through the Responses API.
- In the sales-lead demo, the server responded in under 100 ms. The video says it was not sped up.
- Use case 1: sorting messy free-text user input into sales leads.
- Use case 2: a game where the API reads frames of the road and obstacles and steers the car to the correct lane. It keeps up even at higher speeds, in a fraction of the time a reasoning model would take.
- Use case 3: a split setup where GPT Live 1 handles voice and the Decisions API picks the character's expression while the user talks.
- Design pattern: use a small, fast decision model for routing and classification next to bigger or slower models. This keeps interactions feeling instant.
- The creator says he hasn't tried it, so treat the speed claims as the vendor's own until you test them.
Resources mentioned
- Decisions API · tool · developers.openai.com · paid
A low-latency model API (public beta) that makes fast classification, routing and action choices for apps. - Responses API · docs · developers.openai.com · free
A general model API used as the speed baseline (calling GPT-6 Luna) in the Decisions API comparison. - GPT-6 Luna · tool · community.openai.com · paid
An OpenAI model aimed at fast, high-volume classification and routing.
Also in: Coworker v0.7.0: Open-Source Multi-Bot Workspace with Primary Bot and Web Search (Melvin Vivas on X · notes), DigitalOcean Serverless Inference Now Serves OpenAI GPT-6 Models (Melvin Vivas on X · notes), Codex Orchestrator v0.2.1: GPT-6 Sol orchestrator with Luna subagents (Melvin Vivas on X · notes), DeepSWE results: GPT-6 Sol slightly below GPT-5.6 Sol, but cheaper (Melvin Vivas on X · notes) and 1 more - GPT-Live-1 · tool · openai.com · paid
OpenAI's full-duplex voice model in the API for real-time voice agents that pair with a backend reasoning/tool model.
Also in: GPT-Live-1 Voice Agents in the API, and Charlie Guo (Melvin Vivas on X · notes), GPT-Live-1: OpenAI's Full-Duplex Voice Model for Voice Agents in the API (Melvin Vivas on X · notes)
Try this
- Try the Decisions API public beta and check the under-100 ms latency claim on your own workload.
- Compare it with a general model through the Responses API for routing and classification tasks.
- A tool that sorts free-text inbound messages into sales leads in real time.
- A simple driving game where a decision model reads frames and steers the car between lanes.
- An animated voice character: a real-time voice model handles speech while a decision model picks facial expressions.