Together AI - GoModel

Together AI support is in beta. It is covered by automated tests against a simulated API but has not yet been verified against the live Together AI service. Please report any issues you hit.

Together AI serves open models through an OpenAI-compatible API at https://api.together.ai/v1. GoModel routes these endpoints to it:

GoModel endpoint	Together AI models
`/v1/chat/completions`	chat models (`meta-llama/Llama-3.3-70B-Instruct-Turbo`, `openai/gpt-oss-120b`, …)
`/v1/responses`	chat models, translated through chat completions
`/v1/embeddings`	embedding models (`BAAI/bge-base-en-v1.5`, …)

Configure

Create an API key in the Together AI dashboard and set:

TOGETHER_API_KEY=...

Or in config.yaml:

providers:
  together:
    type: together
    api_key: "${TOGETHER_API_KEY}"
    # base_url defaults to "https://api.together.ai/v1".

Override the base URL or set a model list only when you need to, for example to use Together’s regional endpoints (https://us-api.together.ai/v1):

TOGETHER_BASE_URL=https://api.together.ai/v1
TOGETHER_MODELS=meta-llama/Llama-3.3-70B-Instruct-Turbo,BAAI/bge-base-en-v1.5

TOGETHER_MODELS is a fallback list, used only when /models fails or is empty; CONFIGURED_PROVIDER_MODELS_MODE=allowlist makes it the only list.

Models

GoModel reads Together’s live /models listing, including each model’s type, display name, context length, and per-token prices. Image, audio, video, and rerank models are left out because GoModel does not route those endpoints to Together. Model IDs contain a slash; use them verbatim, or prefix the provider name to pin the route:

curl http://localhost:8080/v1/chat/completions \
  -H "Authorization: Bearer $GOMODEL_MASTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "together/meta-llama/Llama-3.3-70B-Instruct-Turbo", "messages": [{"role": "user", "content": "Hello"}]}'

Use the prefix whenever the ID starts with the name of another configured provider: with an OpenAI provider configured, openai/gpt-oss-120b is looked up on OpenAI and fails, so send together/openai/gpt-oss-120b instead.

Behavior notes

Responses go through chat completions. Together’s native /responses endpoint is a beta that rejects many serverless models, so GoModel translates Responses requests to chat completions, which every chat model supports. With a GoModel response store configured, previous_response_id chains work: the gateway replays the stored history.
Passthrough is opt-in. Add together to ENABLED_PASSTHROUGH_PROVIDERS to expose /p/together/..., which forwards any path under the base URL with the gateway’s Together key. Use it for Together endpoints GoModel does not route, such as images or rerank.

​Configure

​Models

​Behavior notes

Configure

Models

Behavior notes