b stack radar
Together AI logo

Together AI

6.6/10

Together AI is a cloud platform for running, fine-tuning and training open-source AI models. It offers serverless inference, dedicated endpoints and rentable NVIDIA GPU clusters.

Data last updated November 2026 · Our methodology

Overview

Together AI sells AI infrastructure in three layers. The first is serverless inference over a large catalog of open-source models, billed per million tokens. The second is dedicated inference endpoints, billed per GPU-hour. The third is GPU clusters with H100, H200, B200 and newer NVIDIA hardware, which teams rent for training and heavy workloads. Fine-tuning, evaluations and a shared filesystem round out the platform. It suits developers, AI startups and ML research teams that want open models behind an API, or raw GPU capacity, without building a data center. Pricing is pay-as-you-go. Serverless models such as gpt-oss-120B cost $0.15 per million input tokens and $0.60 per million output tokens. Dedicated endpoints start at $5.49 per GPU-hour for an H100. Cluster H100s cost $3.99 per hour on demand, $1.99 preemptible, and $3.45 to $3.69 reserved. Instant Clusters can run Slurm or Kubernetes and are managed through an API, a CLI or Terraform. The company reports a SOC 2 Type 2 examination and says it follows HIPAA requirements. Partners include NVIDIA, Hugging Face, MongoDB, VAST and WEKA. The main limitation is reliability at scale. SemiAnalysis lowered Together to Silver in its ClusterMAX rating, citing frequent complaints from users running 64 or more GPUs. Review sites also report billing and support complaints, and costs can be hard to predict because usage is metered across many services.

Our Rating

6.6
out of 10

AI Software6.8/10

Integration
7
Analytics
5
Customization
8
Ease of Use
6
Support
5
Cost
7
Scalability
8
Security
7

Hosting & Infrastructure6.4/10

Integration
7
Support
5
Cost
7
Customization
8
Security
7
Reliability
5
Ease of Use
6
Analytics
5
Scalability
8

Who Together AI is best for

Best software for Developers

Good fit

  • It is API-first and gives developers OpenAI-style access to many open models plus GPU clusters.

  • Pay-as-you-go token pricing and preemptible GPUs make it cheap for AI startups to get started.

Less suited

  • Reserved clusters and SOC 2 Type 2 help, but SemiAnalysis reports reliability complaints on large clusters.

  • Reviewers say non-technical teams would need engineering help to build anything on it.

Pros & Cons

Pros

  • Low per-token prices on popular open models, for example $0.15 per million input tokens for gpt-oss-120B.
  • Preemptible H100s at $1.99 per GPU-hour cut training costs for jobs that can tolerate interruption.
  • Instant Clusters can be managed by API, CLI or Terraform, with Slurm and Kubernetes options.
  • SemiAnalysis credits its kernel and FlashAttention work with improving performance.
  • 24/7 on-call SRE support for cluster customers.
  • The company reports a SOC 2 Type 2 examination and says it follows HIPAA requirements.

Cons

  • SemiAnalysis reports frequent reliability complaints on clusters of 64 or more GPUs, which led to a downgrade to Silver.
  • Kubernetes clusters lack horizontal node autoscaling, and onboarding needs SSH rather than a downloadable kubeconfig.
  • Trustpilot reviews cite billing and support problems.
  • Usage-metered billing across many services makes monthly costs hard to forecast.
  • It requires ML engineering skill, and reviewers note gaps in the documentation.

Key Features

API and CLI
Managed Kubernetes
Cloud Compute

Not included

  • Managed WordPress
  • Shared Hosting

Together AI Pricing

Dedicated Inference

Contact for pricing
  • Reserved pricing by quote
  • B200 at $9.29 per GPU-hour
  • H100 at $5.49 per GPU-hour
  • H200 at $5.99 per GPU-hour
  • B300 at $11.49 per GPU-hour
  • GB300 NVL72 at $15.79 per GPU-hour
Get Started

GPU Clusters

Contact for pricing
  • Slurm or Kubernetes
  • Reserved discounts for 3 to 90 days
  • B200 at $9.29 per GPU-hour on demand or $4.65 preemptible
  • H100 at $3.99 per GPU-hour on demand or $1.99 preemptible
  • H200 at $5.99 per GPU-hour on demand or $2.99 preemptible
Get Started

Serverless Inference

Contact for pricing
  • Pay per token
  • Some free models, such as Ternary Bonsai 27B
  • gpt-oss-120B at $0.15 input and $0.60 output per 1M tokens
  • Kimi K3.1 Pro at $2.00 input and $10.00 output per 1M tokens
Get Started
Open interactive pricing calculatorGet a tailored cost breakdown for your team size, billing cycle, and add‑ons.

Frequently Asked Questions

You can sign up and start for free, and some serverless models, such as Ternary Bonsai 27B, are listed at no cost. The pricing page does not state a free credit amount.

In GPU clusters an H100 costs $3.99 per GPU-hour on demand, $1.99 preemptible, and $3.45 to $3.69 reserved. On dedicated inference endpoints it costs $5.49 per GPU-hour.

Yes. It announced a completed SOC 2 Type 2 examination in July 2025. It also says it follows HIPAA requirements, including signing BAAs.

Yes. GPU clusters support Slurm and Kubernetes, and Instant Clusters are managed through an API, a CLI or Terraform.

Yes. Supervised and DPO fine-tuning are billed per million tokens, with a minimum charge per job. For example, Llama 3.1 8B costs $0.34 per million tokens with a $4 minimum.

Ranked near Together AI in AI Software

See Together AI in the list

Together AI is #125 in AI Software by our score.

  1. #122
    Baserow logo
    Baserow
  2. #123
    Kolleno logo
    Kolleno
  3. #124
    OneReach.ai logo
    OneReach.ai
  4. #126
    Saturn Cloud logo
    Saturn Cloud
  5. #127
    Gong logo
    Gong
  6. #128
    GitLab logo
    GitLab