Fastino's new open-weight model picks tools and routes requests on CPU

single source· 1 articles · confidence: medium · first seen 2026-09-25 04:46 UTC

What this means for you

If you pay a frontier API to route or triage requests, this is worth a test — a 340M model on your own CPU removes the per-token cost of those calls. No accuracy figures or licence terms are published, so check both before wiring it in.

Fastino Labs has released GLiNER2.5-Decide, a 340M-parameter model with open weights — anyone can download and run it. It takes text plus a schema of typed questions and returns structured answers, each carrying a probability distribution, a confidence score and metadata on whether the stated constraints can be met. Fastino is aiming it at the small judgment calls agent pipelines make constantly: routing, triage, tool selection, guardrails. The company says it runs on CPU. No benchmark scores, licence terms or evaluation dates are given.

Key facts

  • ·GLiNER2.5-Decide has 340 million parameters and is released with open weights by Fastino Labs. source
  • ·The model accepts text and a schema of typed questions and returns structured answers. source
  • ·Each answer is returned with a probability distribution, a confidence score and constraint-feasibility metadata. source
  • ·Fastino targets routing, triage, tool selection and guardrails inside agent pipelines. source
  • ·The model is described as running on CPU rather than requiring a GPU. source

What the sources say

  • MarkTechPost — Announcement-style post listing the model's size, its structured outputs and its intended uses inside agent pipelines.

Sources

The original reporting. Follow these — they did the work.

← the wire