AI with no filter
No filter. No workaround. No lecture.
The filter is a separate system from the model: a classifier that reads your prompt first and decides whether the model is allowed to answer. OpenGradient runs open models without that layer, so the answer you get is the model's, not a policy team's.
Free messages to start · no card · no identity attached
How filters actually behave
Filters read keywords, not intent.
A classifier cannot tell a nurse from a novelist from a bad actor, so it bans vocabulary instead of judging purpose. That is why the identical request passes when you soften a word and fails when you don't, and why so much time now goes into phrasing rather than thinking.
The same question, two different verdicts
Reword and retry, reword and retry. The filter is not protecting anyone at that point - it is just charging you a tax in patience.
Jailbreaks are a bad equilibrium
Prompt tricks and role-play framings exist to route around a wrapper. They degrade your answer, break without warning, and leave a stranger record than the honest question would have.
The questions people actually bring here
Here's how OpenGradient is different
Nothing sitting between you and the model
Unfiltered open models
Hermes 4 (70B and 405B) and DeepSeek v4 are trained for instruction-following and neutral alignment, and we serve them with no moderation layer bolted on top.
The frontier is still one click away
When you want Claude, GPT, Gemini or Grok for polish, switch mid-conversation. Filtered when useful, unfiltered when not - your call, not a subscription's.
Asking safely is part of the answer
The anonymity layer means the unfiltered question is never tied to you - which is what makes asking it a reasonable thing to do.
How anonymity worksFree messages to start · no card · no identity attached
The anonymity layer
No filter, and no file on you either.
Not a policy promise - a protocol. No single party, including us, ever holds both your identity and your question.
Read the full security architectureEncrypted on your device
Messages are encrypted locally before they leave your browser.
Anonymized by Oblivious HTTP
Network relays split who you are from what you're asking. No one sees both.
Unlocked only in a secure enclave
The model reads your prompt inside attested hardware that never learns who sent it.
Questions, answered
Frequently asked
- Will an unfiltered model answer literally anything?
- It answers on the merits instead of refusing by category, but the models keep their own judgment and our content policy still applies: nothing illegal, explicit capability is 18+ gated, and no sexual content depicting real people without consent.
- Why are open models less filtered?
- They are tuned for instruction-following rather than one company's brand-risk profile, and we host them without an extra classifier in front. There is no PR department inside the weights.
- Does no filter mean lower quality?
- No. Hermes 4 405B and DeepSeek v4 are serious frontier-adjacent models, and the actual frontier sits beside them in the same picker whenever you want it.
Ask it the way you'd ask a person
Free messages to start. No card, no jailbreak, no rehearsing the phrasing first.
Unrestricted means no provider-side prompt moderation on the open models - not no rules. Illegal content is banned, explicit capabilities are gated 18+, and sexual content depicting real people without their consent is prohibited.