Together AI

What is Together AI?

Together AI is an AI acceleration cloud that runs open source and custom models for inference, fine-tuning and training. It is made by Together AI and offers serverless model APIs, dedicated endpoints, GPU clusters and post-training services on one platform. The problem it solves is choice and control: teams can use a wide catalogue of models through an OpenAI-compatible API, then move the same workload to reserved capacity or their own fine-tuned model as usage grows.

Key features

• Serverless inference. Pay-per-use APIs cover chat, vision, image, audio, video, transcription, embeddings, reranking and moderation models.
• OpenAI compatibility. The API accepts the OpenAI format at https://api.together.ai/v1, so existing OpenAI SDK code can point at Together with a base URL change.
• Dedicated endpoints. Teams can reserve single-tenant GPU capacity with guaranteed performance, autoscaling and support for custom models.
• GPU clusters. On-demand and reserved clusters include NVIDIA H100, H200, B200 and B300 hardware for training and high-throughput inference.
• Fine-tuning and post-training. The platform supports LoRA fine-tuning, full fine-tuning, direct preference optimisation and managed post-training.
• Code sandbox. A sandbox runs LLM-generated code, with per-vCPU and per-GiB pricing and a separate code interpreter API.

Who is Together AI for?

AI engineering teams that want to run open models without operating GPUs themselves. It suits startups building chat, coding or media features on a usage-based budget, platform teams that need dedicated capacity and predictable throughput, and research groups that fine-tune models and then serve them. Companies that need OpenAI-compatible APIs but want more model choice or lower cost per token are a common fit.

Use cases

• Running a chat or coding assistant on an open model through an OpenAI-compatible API.
• Fine-tuning a model on company data and serving it on a dedicated endpoint.
• Generating images, video, speech or transcriptions from one provider.
• Reserving GPU clusters for training runs or steady production traffic.

Pricing

Together AI is pay as you go and publishes per-model prices. Examples from the pricing page include GLM-5.3 Flash at $0.15 per million input tokens and $0.50 per million output tokens, DeepSeek V4.1 Flash at $0.30 input and $1.20 output per million tokens, and Kimi K3 at $3.00 input and $15.00 output per million tokens. Image generation ranges from about $0.0017 to $0.134 per image depending on the model. On-demand GPU clusters start at $3.99 per hour for an H100, $5.99 for an H200 and $8.19 for a B200. Fine-tuning is billed per million training tokens, and custom deployments are quoted through sales. See the official Together AI pricing page.

Getting started

1. Create an account at api.together.ai and generate an API key.
2. Set TOGETHER_API_KEY in your environment.
3. Install the SDK and send a chat completion request.

pip install together

from together import Together
 
client = Together()
response = client.chat.completions.create(
    model="MiniMaxAI/MiniMax-M3.1-Flash",
    messages=[{"role": "user", "content": "Explain vector databases in two sentences."}],
)
print(response.choices[0].message.content)

Alternatives

• Groq provides fast inference for open models on its own hardware.
• OpenRouter routes requests across many model providers through one API.
• Replicate hosts open source models behind a hosted API.

FAQ

Is Together AI free?

Together AI does not list a free tier on its pricing page. Inference, GPU clusters, fine-tuning and sandboxes are billed by usage, and new accounts can check the dashboard for any current credit offer.

What is Together AI used for?

Together AI is used to run open source and custom models in production. Teams use it for chat, vision, image, audio and video inference, fine-tuning, GPU clusters and code sandboxes.

How much does Together AI cost?

Prices are per model and per resource. For example, GLM-5.3 Flash costs $0.15 per million input tokens and $0.50 per million output tokens. An on-demand H100 GPU costs $3.99 per hour.

Together AI vs Groq?

Both serve open models through an API. Together AI offers a broad catalogue plus dedicated endpoints, GPU clusters and fine-tuning. Groq focuses on low-latency inference on its own hardware. The choice depends on whether you need model variety and training capacity or the fastest possible response time.

Does Together AI support the OpenAI SDK?

Yes. Together AI exposes an OpenAI-compatible endpoint at https://api.together.ai/v1, so the OpenAI SDK can be used by changing the base URL and API key.