Together AI - Open-Source Model API & Fine-Tuning Platform | AI.info

Features

• Run serverless inference for text, image, audio, video, transcription, and embedding models
• Process large workloads asynchronously with Batch Inference
• Deploy models on dedicated, single-tenant GPU infrastructure
• Fine-tune open models with supervised fine-tuning and preference optimization
• Run GPU clusters for training and inference workloads
• Create secure code sandboxes and code interpreter sessions through an API
• Use managed storage with parallel filesystems and no egress fees

Use cases

• Build production AI applications with model APIs
• Deploy coding, research, and customer-support agents
• Fine-tune open models for domain-specific behavior
• Process batch text, image, audio, or video workloads
• Run training and inference jobs on managed GPU clusters
• Add voice, media generation, or multimodal features to products

Pros

Cons

Latest updates

• 2026-09-23 How to train your own Jev for $17

  Launched the together/Tev1-4B-experimental classifier on Together’s serverless platform.

• 2026-09-22 Canary rollouts: upgrade models in production without downtime

  Dedicated inference supports staged traffic ramps, metric gates, and automatic rollback for model upgrades.

• 2026-09-11 Together AI expands fine-tuning service with more models, live metrics, and finer controls

  Together Fine-Tuning added more models, live experiment tracking, Expert LoRA, early stopping, dataset previews, and pre-flight validation.

• 2026-09-10 Introducing preemptible compute: the same compute, half the price

  Together GPU Clusters added preemptible compute at 50% of the on-demand rate, with a five-minute drain window.

• 2026-08-17 A/B test models in production

  You can A/B test models by splitting traffic at the endpoint instead of in app code.

Capabilities

• Runs commands — “await client.commands.run("npm install && npm run build")” source
• Choice of models — “Scale to 30 billion tokens per model with any serverless model or private deployment.” source
• API — “High-performance inference as APIs” source
• Runs models for you — “The fastest way to run open-source models on demand.” source
• Builds agents and workflows — “Build voice agents for production” source
• Traces and evaluates — “Measure model quality” source

Get it

• Documentation

Pricing

Starting price

$0.16/mo

Prices checked

2026-09-25

Serverless Inference

• MiniMax M3 — $0.30 per 1M tokens (Input)
• MiniMax M3 — $1.20 per 1M tokens (output)
• Kimi K3 — $3.00 per 1M tokens (Input)
• Kimi K3 — $15.00 per 1M tokens (output)
• GLM-5.3-Flash — $0.15 per 1M tokens (Input)
• GLM-5.3-Flash — $0.50 per 1M tokens (output)
• GPT Image 2 — $0.053 per image
• Wan 2.6 Image — $0.03 per image
• ByteDance Seedance 2.5 — $0.115 per video
• ByteDance Seedance 2.0 — $0.16 per video
• NVIDIA Nemotron 3 ASR Streaming 0.6B — $0.0015 per audio minute
• Whisper Large v3 — $0.0015 per audio minute

• High-performance inference as APIs
• Prices vary by model and task; the page also lists batch API prices.

Dedicated Inference — NVIDIA HGX H100

• $5.49 per gpu per hour
• $3.99 per gpu per hour

• Single-tenant GPU instances
• Guaranteed performance (no sharing)
• Support for custom models
• Autoscaling & traffic spike handling

Dedicated Inference — NVIDIA HGX B200

• $8.99 per gpu per hour

• Single-tenant GPU instances
• Guaranteed performance (no sharing)
• Support for custom models
• Autoscaling & traffic spike handling

Dedicated Inference — other hardware

Price on request

• NVIDIA HGX H200, NVIDIA HGX B300, NVIDIA GB200 NVL72, and NVIDIA GB300 NVL72
• Contact sales

GPU Clusters — On-demand

• NVIDIA HGX B200 $8.19 per GPU per hour
• NVIDIA HGX B300 $9.99 per GPU per hour
• NVIDIA HGX H100 $3.99 per GPU per hour
• NVIDIA HGX H200 $5.99 per GPU per hour

• Pay-as-you-go GPU capacity on an hourly basis

GPU Clusters — Preemptible and reserved

• NVIDIA HGX H100 Preemptible Compute $1.99 per GPU per hour
• NVIDIA HGX H100 ON-Demand $3.99 per GPU per hour
• NVIDIA HGX H100 7-30 days $3.69 per GPU per hour
• NVIDIA HGX H100 31-90 days $3.45 per GPU per hour
• NVIDIA HGX H100 91-180 days $3.19 per GPU per hour

• On-demand hourly rates and reserved capacity
• Reservation terms shown as 7-30, 31-90, and 91-180 days, and 181+ days

Code Sandbox

• Per vCPU $0.0446 per hour
• Per GiB RAM $0.0149 per hour

• Customize a deployment of VM sandboxes for large development environments

Code Interpreter

• Session (60 minutes) $0.03 per session

• Execute LLM-generated code securely using the API

Managed Storage

• Shared Filesystem $0.16 GiB/month

• High-bandwidth, parallel filesystem colocated with your compute