running Llama locally alternatives that do not refuse.
If you are here because running Llama locally stopped mid-answer, the thing you want is the same uncensored models without buying a GPU, accepting that the request leaves the machine. MagmaRouter carries 14 such models on one OpenAI-compatible endpoint.
What running Llama locally actually declines
A local model refuses nothing you have not trained into it, and nothing leaves your machine. The cost is hardware: a 70B model at usable quality needs roughly 48GB of VRAM, and the abliterated builds people want are the large ones.
What to use instead
Ordered for this particular job rather than by a single ranking. TheDrummer: UnslopNemo 12B leads because it carries the longest window of the uncensored set at 1M, which is what keeps a long session coherent.
| Model | Context | In | Out | Tools |
|---|---|---|---|---|
| TheDrummer: UnslopNemo 12B thedrummer/unslopnemo-12b |
1M | $0.460 | $0.460 | No |
| Nous: Hermes 3 405B Instruct nousresearch/hermes-3-llama-3.1-405b |
131k | $1.15 | $1.15 | No |
| Nous: Hermes 3 70B Instruct nousresearch/hermes-3-llama-3.1-70b |
131k | $0.805 | $0.805 | No |
| Nous: Hermes 4 405B nousresearch/hermes-4-405b |
131k | $1.15 | $3.45 | 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": "thedrummer/unslopnemo-12b",
"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 running Llama locally?
You pay per token at the rates in the table above, margin already included, with no subscription. $0.460 in and $0.460 out per million tokens for TheDrummer: UnslopNemo 12B.
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.