Model checks
Qwen 3.8 27B, seller by seller. Model →
last audit 7 Nov · epoch 30 · evidence checked on-chain2 of 2
sellers answered as the model
How this works (Antseed's announcement ↗)
Approved auditors check that each seller serves the model it sells.
- They buy from it like any buyer: ~100 obscure number questions, compared with the real model.
- The real model misses some too: that is the noise. A seller fails only if it misses clearly more, and two auditors must agree.
- Each pass +2% on $ANTS points, up to +50%. Fails cost more: 1 −4%, 2 −16%, 3 −36%. Checked again every week.
How this works (Antseed's announcement ↗)
Approved auditors check that each seller serves the model it sells.
- They buy from it like any buyer: ~100 obscure number questions, compared with the real model.
- The real model misses some too: that is the noise. A seller fails only if it misses clearly more, and two auditors must agree.
- Each pass +2% on $ANTS points, up to +50%. Fails cost more: 1 −4%, 2 −16%, 3 −36%. Checked again every week.
Passed
2
SAME
Failed
0
DIFF
Not tested
0
the audit could not run
By seller
Compared with qwen/qwen3.8-27b on OpenRouter · pinned to alibaba · reasoning none
| Seller | Differed↑ | × noise | Fail line | Noise | Proof |
|---|---|---|---|---|---|
| JOI SEEDSAME | 9% | 1.1× | 23 under | 8.8% | ↗ |
| Swift AntSAME | 14% | 1.5× | 16 under | 8.8% | ↗ |
Where sellers miss
| Subject | Passed sellers missed | Failed sellers missed |
|---|---|---|
| Chinese literature (publication years) | 6% | — |
| Chinese internet (platform dates) | 13% | — |
| Chinese geography (measurements) | 3% | — |
| Internet culture (dates and numbers) | 17% | — |
| Programming (release years) | 14% | — |
| Biology (chromosome numbers) | 20% | — |
| Chinese history (specific dates) | 5% | — |
| Medical (drug half-lives) | 10% | — |
| Pop culture (box office and ratings) | 5% | — |
| Physics (material properties) | 33% | — |
Audited by approved auditors · evidence on IPFS, hash-checked on-chain.