TypeSafe’s Jev model, a System One decision model, processes typed questions like Choice, Score, or Noul. It operates with 64k context, costs $0.042 per 1M input tokens, and requires a waitlist for access. Developers have identified gaps in Jev, including the lack of image support, self-hosting options, and fine-tuning capabilities.
Cloudflare’s Clef and Clef-flash models, launched on October 1, 2026, offer vision support and are Jev-compatible. Pricing is set at $0.24 and $0.038 per 1M input tokens respectively. OpenAI also entered the space on October 6, 2026, with its Decisions API on gpt-6-luna, priced at $0.10 per 1M input tokens, though it lacks Jev compatibility.
Numerous open-source and specialized alternatives have emerged to address Jev’s limitations. Projects like Standard One, Mapika Decider, and Open-Jev provide various parameter sizes and Apache-2.0 licensing. Some models, such as Fastino’s GLiNER2.5-Decide and Together’s Tev1, utilize different architectures to handle decision-making and span extraction tasks.
Performance benchmarks vary significantly across suites like JevBench and Decision Index. While Jev remains a benchmark for accuracy in several tests, alternatives like Tev1 have demonstrated superior calibration and injection resistance in independent evaluations. Developers are advised to validate these models against their own labeled datasets.
