Decision Models launches a Jev-compatible inference runtime
On 14 October 2026, @airesearch12 (Florian S) launched Decision Models, a hosted inference service for open decision checkpoints on its REAX engine. The launch post advertised models from $0.019 per million input tokens and up to 4× throughput. The current service has Peer-to-Peer and EU tiers; the public /models page lists s1-pro, s1-fast, s1-vision and quyet-1-large, while versions and underlying checkpoints can change.
The service is an inference runtime, not a new trained model. Its evidence page now publishes methods and raw figures and explicitly says the founder runs JevBench, while Decision Models offerings are not listed there. The author also says the work on REAX and DecisionModels will not influence JevBench; this disclosure and the stated independence are relevant context, not independent verification. The launch post’s price and up-to-4× throughput remain UnverifiedClaim unless reproduced from the evidence and measured for the exact model/version.
Why this is not a model entry
- The product is a hosted inference service, not a checkpoint with its own decision-trained weights.
- It exposes a typed, Jev-compatible interface over underlying models; the service itself belongs under Runtimes, outside the quadrant.
- Runtime entry: Decision Models / REAX.