If you are building a product, Mouse-1 is the one you can ship. Torchcast Decision 27B is licensed CC BY-NC 4.0, non-commercial only. On public tool benchmarks the two are within a point (85.6% vs 84.9%), and Mouse-1 leads on phishing detection (72.3% vs 70.4%).
Torchcast Decision 27B, from Torchcast AI, is a LoRA merged into a 27B decision model and published on Hugging Face in early October 2026. Its card reports 33 ms per yes/no question on an H100. It is a strong research model with a licence that keeps it out of commercial products.
At a glance
| Mouse-1 | Torchcast Decision 27B | |
|---|---|---|
| Commercial use | Yes | No (CC BY-NC 4.0) |
| Public tool benchmarks (avg of 4) | 84.9% | 85.6% |
| PhishNChips | 72.3% | 70.4% |
| When2Call | 85.8% | 89% |
| How you run it | Hosted API | Self-host, or Featherless (cold) |
| Calibration | Fitted cutoffs via /v1/calibrations | Board ECE 0.024 |
| Prompt-injection results | Published: 0% flipped | None published |
Public tool benchmarks
Average of BFCL, API-Bank, When2Call and PhishNChips (%) · Higher is better
| Torchcast Decision 27B | 85.6% |
|---|---|
| Mouse-1 | 84.9% |
Function calling, exact match (%) · Higher is better
| Torchcast Decision 27B | 97.7% |
|---|---|
| Mouse-1 | 97.6% |
Next API out of 53 (%) · Higher is better
| Torchcast Decision 27B | 85.2% |
|---|---|
| Mouse-1 | 83.7% |
Answer, call, ask or decline (%) · Higher is better
| Torchcast Decision 27B | 89.0% |
|---|---|
| Mouse-1 | 85.8% |
Phishing or legitimate email (%) · Higher is better
| Mouse-1 | 72.3% |
|---|---|
| Torchcast Decision 27B | 70.4% |
Torchcast leads on When2Call (89 vs 85.8) and API-Bank (85.2 vs 83.7). Mouse-1 leads on PhishNChips. BFCL is a tie within 0.1 points. The board also notes that Torchcast's checkpoint was chosen with the help of index results, which can make its scores here a little optimistic.
How we measured
- Torchcast's scores are from the Jev Decision Index board as of October 20, 2026. Mouse-1 is not on the board; its scores come from our own run of the board's kit on the same rows.
- The licence is from Torchcast's Hugging Face model card, checked October 20, 2026.
- We have not run Torchcast through our agent decision suites.
Sources
Checked October 20, 2026.
See also: Mouse-1 vs Darwin-27B-ZTC · Mouse-1 vs Sieve 27B · All comparisons
FAQ
Can I use Torchcast Decision 27B commercially?
Not under its published licence. Torchcast Decision 27B is released under CC BY-NC 4.0, which does not allow commercial use. Mouse-1 is a commercial API you can build a product on.
Is Torchcast Decision 27B more accurate than Mouse-1?
On the four public tool benchmarks of the Jev Decision Index the two are close: 85.6% for Torchcast against 84.9% for Mouse-1. Torchcast leads When2Call and API-Bank; Mouse-1 leads PhishNChips.
Is Torchcast well calibrated?
Yes. Its expected calibration error on the Decision Index board is 0.024, among the best there. Mouse-1 adds a calibration endpoint that fits a confidence cutoff on your own labelled data.