Models
Models, hosts, and what you pay
Each row is one model on one host. The same model can appear twice when two hosts serve it, and the chat lists them separately. Prices are dollars per million tokens, twice the host's list price, which is what an account pays. A token is about four characters.
| Model | Model Training Company | Company country | US Host Company | Context | Capabilities | Input | Cached input | Output | Inference | Chat | Model card |
|---|---|---|---|---|---|---|---|---|---|---|---|
| GLM 5.3 Flash (Baseten) | Z.ai | China | Baseten | 1,048,576 | Text and images | $0.300 | $0.060 | $1.000 | Not pinned | Offered | Model card |
| DeepSeek V4.1 Flash (Baseten) | DeepSeek | China | Baseten | 1,048,576 | Text and images | $0.600 | $0.014 | $2.400 | Not pinned | Offered | — |
| DeepSeek V4 Flash | DeepSeek | China | Baseten | 1,048,576 | Text | $0.260 | $0.056 | $0.520 | Not pinned | Offered | — |
| GPT-OSS 120B | OpenAI | United States | Baseten | 131,072 | Text | $0.200 | $0.200 | $1.000 | Not pinned | Offered | — |
| Nemotron 3.5 Super VL | NVIDIA | United States | Baseten | 1,048,576 | Text, images, and video | $0.400 | $0.080 | $1.000 | Not pinned | Offered | — |
| Muse Spark Open (Baseten) | Meta | United States | Baseten | 1,048,576 | Text, images, and video | $2.500 | $0.300 | $8.500 | Not pinned | Offered | Model card |
| Beam | Reflection | United States | Baseten | 1,048,576 | Text | $0.800 | $0.080 | $2.400 | Not pinned | Offered | — |
| DeepSeek V4.1 Flash (Together) | DeepSeek | China | Together | 1,000,000 | Text and images | $0.750 | $0.015 | $3.000 | United States | Offered | Model card |
| GLM 5.3 Flash (Together) | Z.ai | China | Together | 1,048,575 | Text and images | $0.375 | $0.075 | $1.250 | United States | Offered | Model card |
| MiniMax M3.1 Flash (Together) | MiniMax | China | Together | 1,048,576 | Text, images, and video | $0.375 | $0.037 | $1.500 | United States | Offered | Model card |
| MiniMax M3 (Together) | MiniMax | China | Together | 1,048,576 | Text, images, and video | $0.750 | $0.150 | $3.000 | United States | Offered | Model card |
| Muse Spark Open (Together) | Meta | United States | Together | 1,048,576 | Text, images, audio, and video | $3.125 | $0.375 | $10.625 | United States | Offered | Model card |
Capabilities are what the host documents. The chat attaches images for a model that reads images, and one video clip for a model marked as reading video. Together's pages also document video or audio for some models; the chat does not send those to Together. Model Training Company is the company that trained the weights. Company country is where that company is based. US Host Company is the US company that serves the model. Inference is separate: it is where that host runs the request. Together rows run on Together's US endpoint and include its 25% rate for that endpoint. Baseten rows use Baseten's shared Model API. Baseten says the regions serving a model on that API can change, unless a compliance policy restricts them, and this server does not set one. A dedicated Baseten deployment can be limited to the United States or the European Union. The chat does not use one of those.
Web search runs through Baseten. A Together model answers without it. See pricing for plans.