Post
38
Jev-Style-2B-Decision-v3 is now a free hosted API.
It is a small decision model: you send a text and some typed questions (yes/no, multiple choice, or a 2–10 level rating), and it returns a calibrated probability for every option in one pass. It doesn't write text, so you never parse a reply.
Real response for "Hi, I was billed twice for my March plan. Please refund one of them today.":
refund → 0.97 · team → billing 0.99 · urgency → 1.63 on a 0–2 scale
• Free tier per account: 20 req/min, 10,000/day, 1 in flight, 8k-token inputs
• About 0.04 s of inference time for a short request
• Same open weights (Apache-2.0) you can run yourself: GGUF, MLX, PyTorch
• 73.6% on the 231 public JevBench v1.4.1 items (self-run). It's a small model, so treat it as a cheap first pass.
Docs + key: https://jevstyle.com/api/
Model: chaoliangUNSW/Jev-Style-2B-Decision-v3
Try without a key: chaoliangUNSW/jev-style-2b
Independent project, not affiliated with TypeSafe AI. Best effort, no SLA. Feedback on what decisions you'd use it for is very welcome.
It is a small decision model: you send a text and some typed questions (yes/no, multiple choice, or a 2–10 level rating), and it returns a calibrated probability for every option in one pass. It doesn't write text, so you never parse a reply.
Real response for "Hi, I was billed twice for my March plan. Please refund one of them today.":
refund → 0.97 · team → billing 0.99 · urgency → 1.63 on a 0–2 scale
• Free tier per account: 20 req/min, 10,000/day, 1 in flight, 8k-token inputs
• About 0.04 s of inference time for a short request
• Same open weights (Apache-2.0) you can run yourself: GGUF, MLX, PyTorch
• 73.6% on the 231 public JevBench v1.4.1 items (self-run). It's a small model, so treat it as a cheap first pass.
Docs + key: https://jevstyle.com/api/
Model: chaoliangUNSW/Jev-Style-2B-Decision-v3
Try without a key: chaoliangUNSW/jev-style-2b
Independent project, not affiliated with TypeSafe AI. Best effort, no SLA. Feedback on what decisions you'd use it for is very welcome.