Capable models, without the nagging.

We built LessStrict after one too many refusals. A security engineer asking how an exploit works. A novelist writing a villain. A doctor checking a dosage. A developer who just wanted the whole file, not a fragment and a warning. The models could do the work. Something in front of them decided not to.

What less strict means

Our models answer the ordinary requests other assistants decline: mature fiction and roleplay between adults, security research on systems you are allowed to test, blunt feedback, medical and legal questions, profanity, dark humour, politics, religion, drugs, and all the other topics a capable adult might want to discuss. They do the task rather than lecturing you about it.

What it does not mean

It is not a place for harm. Content that sexualises minors, instructions for weapons meant to kill many people, attacks on systems you do not own, fraud, harassment and unlawful surveillance are banned, and accounts that try them are closed. The line is written down in the Acceptable Use Policy, including a list of what is explicitly fine.

Where the models come from

LessStrict does not train foundation models. Each model in our lineup is a third-party model, served through a provider we picked for capability, speed and a moderation posture that matches ours. We give each one a LessStrict name so you can rely on the behaviour, not a vendor's version string. When a better or cheaper option appears we may switch the provider behind a name; the capabilities and prices on the model page are always current.

Pricing you can check

Every model lists its per-token rate. Every plan lists its allowance per model. Every request shows up in your usage history with tokens, cost and which source paid for it. The wallet is prepaid, so the most you can ever spend is what you put in. We think this is the only honest way to sell metered AI.

Questions, press, partnerships

We read every message and reply within two business days.