Models
Uncensored open models, sealed in a GPU enclave. Try the live one free, or run any of them in your own Sealed Box.
Uncensored models 1 live · 5 soon
6 open models, freed of their refusals. Each one runs sealed in a GPU enclave.
| Model | Freed by | Sealed Box | Actions |
|---|---|---|---|
| Qwen 3.8 27BLiveOur launch model. Strong reasoning, answers without a lecture. | HereticQwen, freed with Heretic (ARA) | $2.99 /hr−30% $2.09 with the token80 GB GPU | TryRent |
| Gemma 4 31BSoonAmong the best scores on the UGI uncensored leaderboard (65.7). | HereticGoogle, freed with Heretic | $2.99 /hr−30% $2.09 with the token80 GB GPU | SoonRent |
| Gemma 4 26B A4BSoonRefusals cut from 100 in 100 to 11 in 100, published by its author. | HereticGoogle, freed with Heretic v1.2 | $1.99 /hr−30% $1.39 with the token48 GB GPU | SoonRent |
| Qwen 3.8 27BSoonA harder uncensoring pass on the same base model. | AggressiveQwen, uncensored by HauhauCS | $2.99 /hr−30% $2.09 with the token80 GB GPU | SoonRent |
| Qwen 3.8 27BSoonClassic abliteration: the refusal direction removed from the weights. | AbliteratedQwen, abliterated by huihui-ai | $2.99 /hr−30% $2.09 with the token80 GB GPU | SoonRent |
| Dolphin Mistral 24BSoonFine-tuned to follow the user's instructions, not a content policy. | Fine-tunedMistral Small, trained by Dolphin | $1.99 /hr−30% $1.39 with the token48 GB GPU | SoonRent |
Open weights only
Every model here is public. Anyone can download it and check what it is.
Freed, not jailbroken
Refusals are removed from the weights themselves, so there is no prompt trick to wear off.
Sealed when it runs
In the free chat or in your own box, the model answers inside an encrypted GPU enclave.
Unlocked by milestones
Models marked Soon open as the market cap reaches each milestone.