Models

the long tail · 4 of 9 here are hosted nowhere else
$42.10 creditsVerify attestation

Models nobody else hosts

OpenAI, Anthropic, and the major clouds won't carry these. We route them because the providers in the Umbra fleet volunteered, attested, and approved them for hosting. Your prompt still goes to a machine you don't own — the difference is you can verify what ran.

apache-2.0 = permissive model licensellama3.1 = community-attributeda8f4…c01b = model digest-verified (GGUF SHA-256)
try one in playground →

gemma-4-12b-coder-fable5

Q4_K_M128k ctxcode_attested
$0.60/M in
yuxinlu1/gemma-4-12B-coder-fable5…apache-2.0@a8f44d2a8f44d2e9b1c…c01b

A coding fine-tune of Gemma 4 with a Fable5 alignment layer. Nowhere else on the public API market — the provider community ported it because fable5 dropped their upstream support.

hosted by 31 providers

llama-3.3-70b-uncensored

Q4128k ctxhardware
$1.40/M in
NousResearch/Meta-Llama-3.3-70B-I…llama3.1@c01b9f0c01b9f0a4d83…7e29

Abliterated Llama 3.3 70B. Refuses nothing. The popular uncensored build that the major clouds refuse to serve.

hosted by 18 providers

qwen3-32b-roleplay-v2

Q8128k ctxhardware
$1.10/M in
mlabonne/Qwen3-32B-RolePlay-v2-GGUFapache-2.0@4d2e9b14d2e9b1f8c72…a15e

Long-context creative-writing fine-tune of Qwen3. The "warm" tier — 2-3x revenue per slot at Q8 because of the context-length premium.

hosted by 12 providers

mixtral-8x7b-dolphin

Q4_K_M32k ctxhardware
$0.80/M in
mlabonne/Dolphin-Mixtral-8x7B-GGUFapache-2.0@83c0aa783c0aa7e5b91…d4f8

Community Mixtral 8x7B MoE, unfiltered. Slots into 16 GB free on most M-series hosts.

hosted by 9 providers

mistral-7b-claude3

Q4_K_M32k ctxhardware
$0.40/M in
TheBloke/Mistral-7B-Instruct-Clau…apache-2.0@9c2b6e49c2b6e4d2a85…b7c3

Community Mistral 7B distill trained on Claude 3 outputs. Cheap slot-fill; 5.4 GB resident.

hosted by 24 providers

deepseek-coder-v2

Q4_K_M128k ctxhardware
$0.70/M in
TheBloke/deepseek-coder-v2-instru…deepseek@6f81d3a6f81d3a04ce7…e9b2

DeepSeek Coder V2, the coding model that nearly broke SWE-bench. 16B MoE active params.

hosted by 14 providers

yi-34b-abliterated-l3

Q4_K_M200k ctxhardware
$1.00/M in
mlabonne/Yi-34B-Abliterated-L3-GGUFyi@2b5f7c82b5f7c83a1e6…4a91

Yi 34B L3 with the refusal direction ablated. Long-context (200k) for the most-demanding creative tasks.

hosted by 6 providers

llava-1.6-13b-uncensored

Q4_K_Mvisionhardware
$0.90/M in
Andyrasika/llava-1.6-13b-uncensor…apache-2.0@7a3e8d17a3e8d12f0c4…b5a8

LLaVA 1.6 13B vision-language model, refusal ablated. Image + text on attested Apple hardware.

hosted by 4 providers

phi-3.5-mini-uncensored

Q4_K_M128k ctxhardware
$0.20/M in
TheBloke/Phi-3.5-mini-instruct-un…mit@5e9b2f45e9b2f4c8a13…d6e7

Phi-3.5 mini, refusal ablated. 2.3 GB resident — the cheap-slot fill that runs alongside anything else.

hosted by 28 providers