Together AI AI Vendor Risk Profile
AI inference and fine-tuning platform specializing in hosting open-weight models at scale. Provides high-performance inference APIs for Llama, Mistral, and other open models with competitive pricing and developer-friendly tooling.
Risk overview
Risk score: 32/100
Risk tier: Moderate
Safety rating: 68/100 (higher is better)
Lower risk scores indicate lower assessed risk.
Last verified: Apr 19, 2026 Aging
Risk dimensions
| Dimension | Risk score |
|---|---|
| Data Handling | 28/100 |
| IP Exposure | 26/100 |
| Jurisdiction | 8/100 |
| Security | 32/100 |
| Regulatory Compliance | 60/100 |
| Transparency | 70/100 |
| Business Stability | 41/100 |
| Dependency Chain | 32/100 |
| Agent Governance | Not assessed |
Analyst summary
Rating: Acceptable
Together AI is an inference and fine-tuning platform for open-weights models (Llama, Mixtral, Qwen, DeepSeek). It provides SOC 2 Type II, HIPAA capability for qualifying customers, and no-training-on-inputs policy. A strong choice for teams wanting open-weights deployments without self-hosting the infrastructure.
Bottom line: Strong open-weights inference platform; apply model-by-model governance especially on Chinese-origin weights.
Strengths
- Broad open-weights model catalog (Llama, Mixtral, Qwen, Gemma, DeepSeek, etc.)
- SOC 2 Type II certified and HIPAA-capable for qualifying Enterprise customers
- No-training-on-inputs policy with optional zero-retention configurations
- User retains ownership of inputs and outputs
- Competitive pricing for open-weights inference at scale
Concerns
- Serves multiple Chinese-origin open-weights models (Qwen, DeepSeek) that enterprises may need to review separately
- Customer data transits US infrastructure; CLOUD Act exposure applies
- Smaller ecosystem than AWS Bedrock or Azure AI Foundry for deep integration
- Compliance documentation less mature than hyperscaler managed inference offerings
Best for
- Teams standardizing on open-weights models who want managed inference at scale
- Fine-tuning workloads on Llama or Mixtral where cost matters
- ML teams comparing model performance across open-weights options quickly
Avoid if
- Your compliance team rejects Chinese-origin open-weights regardless of where they are hosted
- You need FedRAMP High or sovereign cloud deployment
- You prefer a single-hyperscaler managed AI stack
Citations
- Data Handling — Data Retention Period
Together AI retains API inputs and outputs for up to 30 days for abuse monitoring; Enterprise customers can configure zero data retention.
- Data Handling — Trains On User Data
Together AI does not use customer inputs or outputs to train Together's own models. Inference data is not retained for training.
- Governance — Privacy Policy Url
Together AI Privacy Policy describes collection and processing of user data.
- Governance — Tos Url
Together AI Terms of Service govern use of the inference platform.
- Ip Profiles — User Owns Outputs
Customer retains all rights in Inputs and Outputs. Together AI claims no ownership over customer-generated content.
- Jurisdiction Profiles — Incorporation Country
Together Computer Inc. is a Delaware corporation headquartered in San Francisco, California.
- Security Compliance — Hipaa Compliant
Together AI offers HIPAA BAA for qualifying Enterprise customers on dedicated endpoints.
- Security Compliance — Soc2 Type2
Together AI is SOC 2 Type II and HIPAA-compliant for Enterprise customers.