Nimble

Nimble is pitching its new v4 and v4-max models as a step forward for “decision models,” and it is leaning hard on one simple message: these numbers look better than Jev’s. On the public Decision Index, Nimble says v4-max scores 10.5 points above Jev 1.13, while v4 lands 5.74 points ahead. That is the kind of gap that gets attention fast, especially in a category where small differences can decide whether a model feels useful or merely impressive on a chart.

The company is not stopping at benchmarks, either. It says internal evaluations show 43% fewer mistakes than Jev, plus a 50-point improvement in model routing decisions. Those are the claims that matter more in practice, because routing and error rates are where decision systems either save time or quietly create more work.

Nimble also says the models are available to try for free through its console. That matters because benchmark wins are one thing; getting people to poke at a model themselves is another. Product pages love confidence. Users, less so.

The broader pitch here is familiar: take a model family, frame it as a frontier push, and anchor the story in public scores plus internal evals. Nimble has done that cleanly. Now the question is whether those gains hold up outside its own tests, where the internet is less polite and the mistakes are less forgiving.