Together AI
Cloud platform for open-source model inference, fine-tuning and GPU clusters
Own Together AI? Claim it, free to correct the facts and get the verified badge.
Pricing
Paid
API
Yes
Company
Together AI
Platforms
What is Together AI?
Together AI is a cloud platform for running, customizing and hosting open-source AI models. Developers call chat, vision, image, video, audio, transcription and embedding models through serverless APIs. Billing is per use: per million tokens, per image, per video or per audio minute.
Beyond serverless inference, the platform offers batch inference for large asynchronous jobs, provisioned throughput with a 99% uptime SLA, and dedicated single-tenant GPU instances that can run custom models. On the compute side you can get GPU clusters (H100, H200, B200 and newer), code sandboxes for AI apps and agents, and managed storage with zero egress fees. For model shaping, it runs managed SFT and DPO fine-tuning, custom RL post-training loops and evaluations.
Together AI also publishes systems research such as FlashAttention and ThunderKittens. It suits engineering teams that build AI products on open models and want a path from prototype to production scale.
Key features
- Serverless API for open-source chat, image, video and audio models
- Batch inference scaling to 30 billion tokens per model
- Provisioned throughput with 99% uptime SLA
- Dedicated single-tenant GPU endpoints for custom models
- GPU clusters with H100, H200, B200 and GB300 hardware
- Managed SFT and DPO fine-tuning plus custom RL post-training
- Code sandboxes for AI apps and agents
- Managed storage with zero egress fees
Use cases
Who it's for
Adding open models to an app
Call models like GLM, DeepSeek, Qwen or Kimi through a serverless API and pay per token, with no infrastructure to manage.
Processing large offline workloads
Run asynchronous jobs through the Batch API to classify, summarize or generate data at lower cost.
Customizing a model
Fine-tune open-source models with SFT or DPO, or define your own RL loop, then measure quality with evaluations.
Training at scale
Rent GPU clusters, from self-serve instant clusters up to thousands of GPUs, for training and research.
Together AI pricing
From the vendor's site, checked Nov 8, 2026; we re-check prices monthly. Prices change — confirm on their pricing page.
Serverless Inference
usage-based
From $0.00 per 1M tokens (e.g. gpt-oss-120B $0.15 input / $0.60 output)
- Pay per token, image, video or audio minute
- Chat, vision, image, audio, video, transcription models
- Batch API pricing available
Provisioned Throughput
usage-based
Per PTU per minute
- Reserved capacity in throughput units
- 99% uptime SLA
- Cost estimator on site
Dedicated Inference (On-demand)
hourly
$5.49/GPU/hour (H100), $8.99/GPU/hour (B200)
- Single-tenant GPUs
- Custom model support
- Autoscaling
Dedicated Inference (Reserved)
contract
Custom
- Contact sales
Get an email when Together AI changes its prices
One mail per change, a stop link in each. Nothing else.
How Together AI compares
- Together AI has no free plan, while 1 of the 3 alternatives below has one.
| Tool | Starts at | Free plan | Free trial | API | Open source | |
|---|---|---|---|---|---|---|
| TTogether AI | — | — | — | — | ||
F Fireworks AI | — | — | — | — | Together AI vs Fireworks AI | |
G Groq | — | — | — | — | — | Together AI vs Groq |
O OpenRouter | — | — | — | Together AI vs OpenRouter |
The closest alternatives in our directory by category. “Starts at” is the cheapest paid plan per month, as each vendor lists it.
Pros and cons
Strengths
- Wide catalog of open-source models across text, image, video and audio
- Transparent per-token and per-GPU-hour prices
- Some models listed at $0.00 per 1M tokens
- Covers inference, fine-tuning and compute in one platform
Limitations
- Reserved GPU pricing requires contacting sales
- Usage-based billing makes monthly costs harder to predict
- Aimed at developers; requires API integration
Together AI FAQ
How is Together AI priced?
Usage-based: serverless models are billed per 1M tokens, per image, per video or per audio minute. Dedicated GPUs are billed per GPU hour, e.g. $5.49 for H100 on-demand.
Are there free models?
Yes. Some models, such as Ternary Bonsai 27B, are listed at $0.00 per 1M tokens.
Can I deploy my own model?
Yes. Dedicated Model Inference and Dedicated Container Inference support custom models on single-tenant GPUs.
Does it support fine-tuning?
Yes. It offers managed SFT and DPO fine-tuning jobs, custom RL/SFT post-training loops, and evaluations.
Is there an uptime SLA?
Provisioned Throughput includes a 99% uptime SLA.
Used Together AI? Rate it
Sign in to rate this tool and share how you use it.
Together AI
Paid