MODELS
Every model here is published by its original maker under a licence that allows commercial use. Generated from what the gateway serves: if it is listed, you can call it with "model": "<id>".
25 MODELS
| ID | MODEL | PUBLISHER | LICENCE | INPUT | SIZE | CONTEXT | COLD START | PUBLISHER'S CARD | GPU | PRICE WHILE RUNNING |
|---|---|---|---|---|---|---|---|---|---|---|
mistral-nemo-12b | Mistral Nemo Instruct 2407 | Mistral AI | Apache-2.0 | Text | 12B | 32k | about 44 s | 1 x 24 GB GPU | from $0.22 / h | mistralai/Mistral-Nemo-Instruct-2407 |
mistral-small-32-24b | Mistral Small 3.2 24B Instruct | Mistral AI | Apache-2.0 | Text | 24B | 64k | about 145 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | mistralai/Mistral-Small-3.2-24B-Instruct-2506 |
qwen25vl-7b | Qwen2.5-VL 7B Instruct | Qwen (Alibaba Cloud) | Apache-2.0 | Text + images | 7B | 32k | about 82 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | Qwen/Qwen2.5-VL-7B-Instruct |
qwen3vl-2b | Qwen3-VL 2B Instruct | Qwen (Alibaba Cloud) | Apache-2.0 | Text + images | 2B | 32k | about 62 s | 1 x 24 GB GPU | from $0.22 / h | Qwen/Qwen3-VL-2B-Instruct |
qwen3vl-8b | Qwen3-VL 8B Instruct | Qwen (Alibaba Cloud) | Apache-2.0 | Text + images | 8B | 32k | about 38 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | Qwen/Qwen3-VL-8B-Instruct |
qwen38-27b-base | Qwen3.8 27B | Qwen (Alibaba Cloud) | Apache-2.0 | Text | 27B | 256k | about 64 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | Qwen/Qwen3.8-27B |
qwen36-35b-a3b | Qwen3.6 35B-A3B | Qwen (Alibaba Cloud) | Apache-2.0 | Text | 35B MoE (3B active) | 256k | about 50 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | Qwen/Qwen3.6-35B-A3B |
qwen35-9b | Qwen3.5 9B | Qwen (Alibaba Cloud) | Apache-2.0 | Text | 9B | 128k | about 55 s | 1 x 24 GB GPU | from $0.22 / h | Qwen/Qwen3.5-9B |
qwen25-14b | Qwen2.5 14B Instruct | Qwen (Alibaba Cloud) | Apache-2.0 | Text | 14B | 32k | about 176 s | 1 x 24 GB GPU | from $0.22 / h | Qwen/Qwen2.5-14B-Instruct |
gemma4-31b | Gemma 4 31B it | Apache-2.0 | Text + images | 31B | 64k | about 74 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | google/gemma-4-31B-it | |
gemma4-26b-a4b | Gemma 4 26B-A4B it | Apache-2.0 | Text + images | 26B MoE (4B active) | 32k | about 51 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | google/gemma-4-26B-A4B-it | |
gemma4-e4b | Gemma 4 E4B it | Apache-2.0 | Text | E4B | 64k | about 95 s | 1 x 24 GB GPU | from $0.22 / h | google/gemma-4-E4B-it | |
llama31-8b | Llama 3.1 8B Instruct | Meta | Llama 3.1 Community Licence | Text | 8B | 32k | about 69 s | 1 x 24 GB GPU | from $0.22 / h | meta-llama/Llama-3.1-8B-Instruct |
gemma4-12b | Gemma 4 12B Instruct | Apache-2.0 | Text + images | 12B | 32k | about 106 s | 1 x 24 GB GPU | from $0.22 / h | google/gemma-4-12B-it | |
laguna-xs | Laguna XS 2.1 | poolside | OpenMDW-1.1 | Text | 33B | 64k | about 67 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | poolside/Laguna-XS-2.1 |
nemotron3-30b-a3b | Nemotron 3 Nano 30B-A3B | NVIDIA | NVIDIA Open Model License | Text | 32B (3B active) | 32k | about 62 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 |
deepseek-r1-qwen-7b | DeepSeek-R1-Distill-Qwen-7B | DeepSeek | MIT | Text | 7B | 128k | about 242 s | 1 x 24 GB GPU | from $0.22 / h | deepseek-ai/DeepSeek-R1-Distill-Qwen-7B |
deepseek-r1-qwen-14b | DeepSeek-R1-Distill-Qwen-14B | DeepSeek | MIT | Text | 14B | 64k | about 64 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | deepseek-ai/DeepSeek-R1-Distill-Qwen-14B |
deepseek-r1-qwen-32b | DeepSeek-R1-Distill-Qwen-32B | DeepSeek | MIT | Text | 32B | 32k | about 55 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | deepseek-ai/DeepSeek-R1-Distill-Qwen-32B |
deepseek-r1-llama-8b | DeepSeek-R1-Distill-Llama-8B | DeepSeek | MIT (base model: Llama 3.1 Community License) | Text | 8B | 32k | about 48 s | 1 x 24 GB GPU | from $0.22 / h | deepseek-ai/DeepSeek-R1-Distill-Llama-8B |
deepseek-r1-0528-qwen3-8b | DeepSeek-R1-0528-Qwen3-8B | DeepSeek | MIT | Text | 8B | 32k | about 40 s | 1 x 24 GB GPU | from $0.22 / h | deepseek-ai/DeepSeek-R1-0528-Qwen3-8B |
phi-4 | Phi-4 | Microsoft | MIT | Text | 14B | 16k | about 39 s | 1 x 48 GB GPU (A40 class) | from $0.46 / h | microsoft/phi-4 |
phi-4-mini | Phi-4-mini-instruct | Microsoft | MIT | Text | 3.8B | 64k | about 48 s | 1 x 24 GB GPU | from $0.22 / h | microsoft/Phi-4-mini-instruct |
phi-4-mini-reasoning | Phi-4-mini-reasoning | Microsoft | MIT | Text | 3.8B | 64k | about 44 s | 1 x 24 GB GPU | from $0.22 / h | microsoft/Phi-4-mini-reasoning |
gpt-oss-20b | gpt-oss-20b | OpenAI | Apache-2.0 | Text | 21B MoE (3.6B active) | 128k | about 118 s | 1 x 24 GB GPU | from $0.22 / h | openai/gpt-oss-20b |
CONTEXT is the model's window here; your level caps each request (see pricing). COLD START is the measured median time to the first token when the model has to start; once running it stays up 10 minutes after the last request. Quantised weights are standard quantisations of the publisher's release; the publisher's licence applies to your use of each model.