Together Inference

What it does

Together Inference's serverless option takes API requests and returns generated output from hosted open-weight models. Inputs can be chat messages, images, audio, code prompts or text for embeddings; outputs include text, images, video, speech and transcripts. Models include GLM, Qwen, DeepSeek, Kimi, MiniMax and gpt-oss families, so there is nothing to deploy or manage.

Developers call it through an OpenAI-compatible API: set a Together API key and the base URL, then use existing OpenAI Python or TypeScript clients or cURL. Model IDs follow a provider/model naming pattern, so OpenAI model names return errors. Together describes speed work such as adaptive speculative decoding and quantization, and publishes its own benchmark comparisons. Teams can move to provisioned throughput, dedicated endpoints or custom containers for steadier traffic or their own models.

It suits developers and teams building applications on open models who want usage-based billing without managing GPUs. An API key from the Together console is required. Assistants, Threads, Runs and moderation endpoints are not implemented, the OpenAI-shaped batch and files APIs are not exposed, and some parameters (such as seed, logit_bias and service_tier) behave differently or are ignored. Speed claims are the vendor's own.

What it can do

• Serverless chat, vision and code inference over open-weight models
• Image, video, text-to-speech and speech-to-text generation via one API
• OpenAI-compatible endpoints for chat, embeddings, images and audio
• Function calling and structured outputs
• Batch API pricing for supported models
• US-only geo-specific inference endpoint on select models

Works with

• OpenAI Python SDK
• OpenAI TypeScript SDK

Example workflows

Switch an existing app to open models

Input: an app using the OpenAI Python client. Steps: create a Together API key, set base_url to https://api.together.ai/v1, change the model to a Together model ID. Output: chat completions from an open-weight model.

Generate marketing images

Input: text prompts and a chosen image model. Steps: call the image generation endpoint with the model ID and prompt. Output: image URLs or base64 images billed per image.

Before you choose

• OpenAI model names return 404; use Together provider/model IDs
• Assistants, Threads, Runs and moderations are not supported, and some parameters are ignored
• Speed comparisons come from Together's own page; verify for your workload

Listed under

• Coding
• Sandboxes and infrastructure
• Fine-tuning
• GPU cloud
• Image generation
• Inference API
• LLM API
• Open models

Get started

Sign in at the Together console

1. Create an account and generate an API key in the Together console
2. Set TOGETHER_API_KEY and point your OpenAI-compatible client at the Together base URL
3. Pick a model ID from the models list and send a test request

Serverless text models are priced per 1M tokens with separate input, cached and output rates; image, video and audio are priced per unit. See the pricing page for the model rate table. A US-only endpoint costs 1.25x standard rates. Provisioned throughput and dedicated container rates are not published here.

Source: together.ai/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Serverless Inference | Together AI ↗www.together.ai
• Pricing | Together AI ↗www.together.ai
• OpenAI compatibility - Together AI docs ↗docs.together.ai

Company website

together.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.