Fastino Labs released two open-weight models for finance and healthcare, post-trained entirely by an autonomous agent on NVIDIA’s Nemotron 3.5 Lightning base.
On 11 August 2026, Palo Alto-based applied AI research lab Fastino Labs (creators of the GLiNER model family) announced two domain-specific open-weight models: Fastino-Nemotron-3.5-Lightning-Finance and Fastino-Nemotron-3.5-Lightning-Healthcare.
Both models were post-trained on NVIDIA’s newly released Nemotron 3.5 Lightning, a 30-billion parameter Mixture-of-Experts model with 3 billion active parameters. Both models are available immediately on Hugging Face under the permissive Apache 2.0 open-source licence.
Both models were post-trained entirely using the Fastino Fine-Tuning Agent, an autonomous autoresearch agent that handles task research, data curation, evaluation set creation, parallel training runs, error recovery, data contamination tests, and final checkpoint selection from plain language prompts.
The agent completed the full post-training workflow for both models in less than a day, a process that typically requires weeks for human post-training teams. The Fastino Fine-Tuning Agent is available as a private preview and will see a full general release supporting various open-weight models in the coming weeks.
For the finance model, FinQA execution accuracy lifted performance from 15.86% (base model) to 59.23% (+43.37 percentage points), while TAT-QA (F1) increased from 19.01% to 56.63% (+37.62 percentage points). It outperformed the base model across five distinct financial reasoning benchmarks (including SEC-Num, FinEntity, and BizFinBench).
Fastino Labs’ healthcare model demonstrated gains across eight medical benchmarks, including verified improvements on MEDEC, MedCalc-Bench, and HealthBench (improving HealthBench from 40.62% to 48.17% across 700 reserved clinical conversations). Both models demonstrated capability transfer to unseen related tasks, confirming generalised domain proficiency rather than benchmark overfitting.
