DecisionModels launches decision API with probability outputs, claims 4.5x GPU throughput and half of Jev's price
airesearch12 · x · 2026-10-15
- airesearch12 explains the branding behind DecisionModels: the community has settled on calling these small inference models "decision models" (some prefer "system one models"), so he registered the main domain plus 35 related names.
- The product is a "Jev-compatible" decision API for open-weights models: one answer per option with probabilities, aimed at ticket routing, action approval and draft scoring; one API covers s1-fast, s1-pro and s1-vision.
- Self-measured claims (method and raw data public): s1-pro delivers 2.6x the throughput of its reference server, Quyet-1.0-Large 4.5x on the same GPU; s1-fast priced at half of Jev.
- The model behind s1-pro, vLLM-SR d3, tops the community Decision Index 0.3.2 board with 66.26 (Oct 12, 2026).
- Two tiers: a global peer-to-peer tier using spare EU capacity plus rented GPUs from independent providers (who see request content in cleartext, opt-in only), and an EU tier built for GDPR and EU AI Act compliance.
More from Infra
- Routing GPT 5.1 medium tasks to open models with Jev cuts costs ~73% at 98.4% success — gaganghotra_ · 2026-10-15
- EU-China talks aim to save ASML supplier Sindlhauser from export-control squeeze — teortaxesTex · 2026-10-15
- Why Local Echo Cancellation Still Matters Even With Echo-Robust gpt-live — juberti · 2026-10-15
- Jensen Huang Says He Won't Sell GPUs to Irresponsible AI Labs — rohanpaul_ai · 2026-10-15
- aiand_ CEO Talks Scaling Japan's Inference Infrastructure at Nikkei GenAI Summit — DavidBennett__ · 2026-10-15
- Citi sees agentic CPU TAM growing 14X to $157.4B by 2030, over half of total CPU market — Beth_Kindig · 2026-10-15