Model cards
Model cards
One card per Sieve model, with the base model, limits, calibration, benchmarks and licence.
Sieve models are decision models. They return a probability for each option of a typed question and never generate text. Pick one by name when you call the API, or use sieve-auto to let the cascade choose.
Read the numbers carefully
Benchmark numbers are self-reported by the model authors unless a card says they come from a public leaderboard. Latency was measured on different GPUs and setups, so do not compare it across cards. The API names below follow the current model catalog; check GET /v1/models for what your key can use.
| Model | API name | Summary |
|---|---|---|
| Sieve-0.8B | sieve-0.8b | A small decision model |
| Sieve-27B-v2 | sieve-27b-v2, alias sieve-27b | A large fully fine-tuned decision model with per-category calibration |
| Sieve-27B | sieve-27b-v1 | A large decision model |
| Sieve-2B-v3 | sieve-2b-v3, alias sieve-2b | A small decision model with per-category calibration |
| Sieve-2B | Not served | A small decision model |
| Sieve-31B | sieve-31b | The accuracy-tier decision model |
| Sieve-4B | sieve-4b | The hosted low-cost decision model |
| Sieve-9B-v2 | sieve-9b-v2, aliases sieve-9b and sieve-latest | A mid-size decision model with per-category calibration |
| Sieve-9B | sieve-9b-v1 | A mid-size decision model |
| Sieve-Vision-4B | Not available yet | A vision decision model for multiple-choice questions about images |
All cards are also published on Hugging Face. See Question types and Confidence for how to use the output.
Next steps
Last updated