Bespoke Labs drops Nimble-v4 decision models

Routing prompts between different AI models is tricky, and Bespoke Labs thinks it has a smaller, smarter way to do it.

A 9B model takes the decision crown

Bespoke Labs rolled out Nimble-v4 and Nimble-v4-max, claiming the top spot on the Decision Index public benchmark.

Going head-to-head with Jev 1.13, Nimble-v4-max scored plus 10 and Nimble-v4 scored plus 6. Internal evaluations show the new pair made 43% fewer mistakes than Jev and delivered a 50 percentage point leap on model routing decisions.

Notably, the 9B-parameter Nimble-v4 punched above its weight against several larger models, standing out in coding tasks, fact-checking, and chess. Both models are already live through the company's API console.

Why it matters

If you build multi-agent setups or router pipelines, these models are worth testing immediately. Rather than throwing every query at an expensive frontier LLM, a dedicated 9B router can direct traffic at much lower latency and cost.

We think focusing on decision-making rather than raw generative bloat is a sharp move. That said, routing benchmarks often look cleaner than messy production data, so its real value will depend on how reliably it navigates ambiguous user prompts without tripping up.