Together AI
Pay-per-token cloud for serverless inference, fine-tuning, and GPU clusters across 200+ open-source and frontier AI models.
Last verified:
Together AI is a San Francisco-based AI cloud platform offering serverless pay-per-token inference, dedicated single-tenant endpoints, fine-tuning, and on-demand/reserved GPU clusters across 200+ open-source and frontier models (Llama, DeepSeek, Qwen, FLUX, and more). It exposes an OpenAI-compatible REST API with official Python and TypeScript SDKs, plus a documentation-search MCP server for coding agents. Founded in 2022, Together raised an $800M Series C in July 2026 with backing from NVIDIA, Salesforce Ventures, and General Catalyst, alongside new GPU capacity commitments.
Pricing
| Tier | Price | Key features |
|---|---|---|
| Serverless Inference | Custom | Chat/text models: $0.05-$1.74 per 1M input tokens, $0.15-$4.50 per 1M output tokens, Image generation: $0.003-$0.134 per image, Video generation: $0.14-$3.20 per video, Embeddings from $0.02 per 1M tokens; cached-token discounts on some models |
| Provisioned Throughput | $0.05/usage | Fixed-capacity reservation, Estimated up to 83% savings vs. pay-per-token at high sustained volume |
| Dedicated Endpoints | $5.49/usage | H100: $5.49/hr, B200: $8.99/hr, H200/B300/GB200/GB300: contact sales |
| GPU Clusters | $3.99/usage | On-demand H100 $3.99/hr, H200 $5.99/hr, B200 $8.19/hr, Reserved (7-180+ day) discounts down to ~$3.09/hr H100, Kubernetes/Slurm support |
| Fine-Tuning | $0.48/usage | Up to 16B params: $0.48 SFT / $0.54 DPO per 1M tokens, 70-100B params: $2.90-$3.20 per 1M tokens, $4.00 minimum charge per job |
| Enterprise / Custom Hardware | Custom | Premium GPUs (H200, B300, GB200/GB300 NVL72) via contact sales, Custom private deployments and SLAs |
Source: https://www.together.ai/pricing
Agent-readiness
- Discoverability
- 65/100
- Machine-readable content
- 75/100
- Programmatic access
- 90/100
- MCP / agent support
- 55/100
- Self-serve transactability
- 85/100
- Transparency
- 70/100
Security & compliance
- SSO / SAML
- Not publicly documented
- Encryption at rest
- Not publicly documented
- Encryption in transit
- Yes
- Audit log
- Not publicly documented
- Pentest report
- Not publicly documented
"Not publicly documented" means Crail found no public evidence either way — not a confirmed absence. See our methodology. Trust center: https://trust.together.ai.
FAQ
Does Together AI have a free tier?
No — Together AI does not offer a free tier as of 2026-07-27.
Can you buy Together AI without talking to sales?
Yes, Together AI can be purchased self-serve without a sales call.
Does Together AI have an official MCP (Model Context Protocol) server?
Yes, Together AI publishes an official MCP server.
Reviews
No reviews yet for Together AI. Be the first to submit a verified review.
Methodology & disclosure
Data verified via agent crawled, last checked 2026-07-27. Agent-readiness methodology version crail-ar-v0.1. No vendor payment influences these scores — see our methodology page.