Fastino Labs Launches Pioneer for Automated Fine-Tuning and Adaptive Inference of Language Models

Fastino Labs has launched Pioneer, a language model fine-tuning agent and adaptive inference platform for open source small language models, announced in a press release. Pioneer allows developers to fine-tune and deploy models such as Qwen, Gemma, Llama, Nemotron, and GLiNER through a single prompt.

Pioneer introduces adaptive inference, a model serving method where deployed models continuously retrain on live production data without human input. The platform automatically validates and promotes improved checkpoints, maintaining performance gains over time. In benchmark tests simulating deployment drift, Pioneer achieved consistent improvement while traditional retraining methods degraded.

The platform includes two modes. Agent Mode enables users to fine-tune and deploy models in minutes through a chat interface, while Deep Research Mode executes fully autonomous fine-tuning with web browsing, data discovery, and iterative optimization. Fastino also introduced AdaptFT-Bench, a benchmark for evaluating autonomous model improvement in production conditions.

According to Fastino, Pioneer achieved accuracy gains of up to 83.8 percentage points over base models, completing end-to-end runs in about six hours at a cost of roughly 35 dollars. The company has raised 25 million dollars in funding from investors including Khosla Ventures, Insight Partners, M12, NEA, and Valor Equity Partners.