ChatGPT alternatives that do not refuse.

If you are here because ChatGPT stopped mid-answer, the thing you want is a model that will discuss exploitation, malware behaviour or adult fiction without a refusal in the middle of the answer. MagmaRouter carries 14 such models on one OpenAI-compatible endpoint.

What ChatGPT actually declines

OpenAI's usage policies prohibit using its models to create malware, to facilitate weapons development, or to generate sexual content involving minors, and the deployed assistant declines a much wider range of security and adult topics than the policy strictly requires.

Worth being straight about Nothing here is a jailbreak, and none of it makes an illegal act legal. A loosely aligned model answers more questions; it does not change what you are allowed to do with the answer, and it is frequently more confidently wrong than the model you left.

What to use instead

Ordered for this particular job rather than by a single ranking. Sao10K: Llama 3 8B Lunaris leads because it is the cheapest of the uncensored set on output at $0.057 per million tokens.

ModelContextInOutTools
Sao10K: Llama 3 8B Lunaris
sao10k/l3-lunaris-8b
8k $0.046 $0.057 No
MythoMax 13B
gryphe/mythomax-l2-13b
8k $0.092 $0.127 No
TheDrummer: UnslopNemo 12B
thedrummer/unslopnemo-12b
1M $0.460 $0.460 No
TheDrummer: Cydonia 24B V4.1
thedrummer/cydonia-24b-v4.1
131k $0.345 $0.575 No

Switching takes one base URL

Anything that speaks the OpenAI chat completions format works unchanged. Point it here, use a MagmaRouter key, and name the model from the table.

curl https://api.magmarouter.com/v1/chat/completions \
  -H "Authorization: Bearer rl-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sao10k/l3-lunaris-8b",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

FAQ

Is MagmaRouter actually uncensored, or does it just say so?

MagmaRouter does not moderate requests. It prices the call, holds the money against your key cap and passes the request to the model. What comes back is whatever that model does: an uncensored finetune refuses far less than a frontier model but can still decline, and only an abliterated build has the refusal direction removed outright. Neither tag is a promise that any given answer will arrive.

What does it cost compared with ChatGPT?

You pay per token at the rates in the table above, margin already included, with no subscription. $0.046 in and $0.057 out per million tokens for Sao10K: Llama 3 8B Lunaris.

Are my prompts stored?

No. The database keeps the model id, token counts and cost. It does not keep the prompt or the completion, and every call returns a signed receipt of the prompt hash so you can prove afterwards what you sent.

Can I cap what a key is allowed to spend?

Yes, and the cap is enforced before the request goes upstream rather than reconciled afterwards. The reply cannot physically cost more than the hold placed before it was sent.

Related