TypeSafe AI has launched Jev, a System One model developed after two years in stealth. The model processes states and questions in parallel, utilizing a training method known as Reinforcement Learning for Calibrated Decisions. It is priced at $0.042 per million input tokens, with no cost for output, and features a 32K context window.
Performance benchmarks show Jev achieving accuracy comparable to Sonnet 5 while maintaining lower latency. In specific tasks, the model scored 76.0% in customer service and 61.8% in invoice processing. According to Vercel, Jev achieved the fastest adoption rate in AI Gateway history, with 13% of paid teams integrating it within 24 hours of release.
The model is being applied in various technical workflows, including contract interpretation at the network edge and as a replacement for expensive LLM rerankers. Research indicates that Jev reduced median decision latency by 22.4% compared to DeepSeek and 61.9% against Gemini. Additionally, a demonstration running the model at 10 queries per second estimated costs at $7 per hour.
Following the launch of Jev, Fastino Labs introduced competing models, GLiDE and GLiNER2.5-Decide. The market for decision-focused AI also includes open-source reproductions and experimental integrations from platforms like Arize and Langfuse. These tools are designed for scenarios requiring bounded answers that trigger code branching, rather than human-readable text generation.
