DecisionModels Launches Open-Weight Decision Model API at $0.019 per Million Input Tokens

JevBench author airesearch12 announced on October 15 the launch of DecisionModels, a platform focused on inference services for "decision models" (also called system one models), priced at $0.019 per million input tokens. This isn't a chat model—it's a production-oriented decision API that outputs an answer plus a probability for each option.

Confirmed

• Positioning: A single "decision API" covering models such as s1-fast, s1-pro, and s1; it doesn't generate free-form text, but returns an answer and probability per option for use cases like routing tickets, approval actions, and scoring drafts
• Underlying mechanism: The platform is powered by REAX (Bittensor subnet 92); anyone with a GPU can host these open-weight decision models, and the AI engineering community is incentivized to improve the REAX high-speed inference engine
• Performance claims: airesearch12 says that over the past four weeks they benchmarked multiple decision models on their own JevBench and launched the best performers; the company claims up to 4x speed on its REAX engine and 4.5x the throughput of the reference server
• Brand strategy: airesearch12 said the industry has settled on calling these small inference models "decision models," and he has registered the main domain plus roughly 35 related domains

Why it matters

• Decision models separate classification/routing tasks from generative LLMs, serving production scenarios at lower cost and higher throughput—the $0.019 per million token pricing is quite competitive
• Using a Bittensor subnet to incentivize community GPU hosting is an experiment in decentralized inference; if the performance claims hold, it could influence pricing and deployment of traditional cloud small-model APIs

Unconfirmed

• The "4x speed" and "4.5x throughput" figures come solely from operator airesearch12 with no third-party verification yet; there's also a potential conflict of interest in benchmark selection, as the author also runs JevBench

2026-10-15 ~ 2026-10-15 · 5 related posts

• [source] DecisionModels launches Jev-compatible decision API: open models, 4x faster, from $0.019/1M tokens — airesearch12 · 2026-10-15
• [source] DecisionModels runs on Bittensor subnet 92, incentivizing community GPU hosting — airesearch12 · 2026-10-15
• [source] Decision Models launches P2P decision API returning option probabilities, 2.6x throughput — airesearch12 · 2026-10-15
• DecisionModels launches decision API with probability outputs, claims 4.5x GPU throughput and half of Jev's price — airesearch12 · 2026-10-15
• DecisionModels launches: open decision models up to 4x faster, $0.019 per 1M tokens — airesearch12 · 2026-10-15