Skip to main content
Together AI−49 Negative

GILD Review

GILD's call and the buying facts behind it.

From GILD research

Should we adopt Together AI?

Which Together AI inference mode fits each workload, and which open models will your app depend on?

  1. 30-Second TakePick Together AI if you want open models on an OpenAI-compatible API without running GPUs.
  2. Best fitAI and platform teams that run open-weight models through an OpenAI-compatible API, start on prepaid per-token serverless, and move steady traffic to provisioned throughput or dedicated endpoints.GILD Review
  3. Consider another workflowIf prompts must not be stored by default, compare Fireworks AI, whose data-handling page states Zero Data Retention by default. If you need an organization-wide audit log of user and admin actions, ask Together in writing, since its documented audit events cover dedicated endpoints.GILD Review

Drawn from 46 sources and 3 operator insights

GILD analysis

Together AI Review

Developers pick Together AI for open models billed per token on a host that runs the models itself, and report tool-calling errors, retired serverless models and credit problems. Accounts are fully prepaid from a US$15 minimum, and provisioned throughput or dedicated endpoints reserve capacity when traffic is steady. Adopt it with tool calls tested per model, production pinned to models Together serves serverless, and a fallback host.

  • Inference Hosting
  • AI Costs

The decision

Best fit
AI and platform teams that run open-weight models through an OpenAI-compatible API, start on prepaid per-token serverless, and move steady traffic to provisioned throughput or dedicated endpoints.
Consider another workflow
If prompts must not be stored by default, compare Fireworks AI, whose data-handling page states Zero Data Retention by default. If you need an organization-wide audit log of user and admin actions, ask Together in writing, since its documented audit events cover dedicated endpoints.
Adoption condition
Tool calls and long prompts tested on each model you pin, only models Together serves serverless in production, credits funded ahead of use, and a Scale or Enterprise account if you need SSO.

Basis: Together AI's documentation, pricing, trust center and SOC 2 post, stored on Nov 1, 2026, 41 public practitioner accounts, and three named deployment accounts from Ai2, MLCommons and GoodParty.org. GILD ran no benchmark of its own.

Buying & deployment factsWhat a team building on it should verify before production traffic grows.
How you buy it

US$15 minimum

Fully prepaid. API access stops at a zero balance, and provisioned throughput goes through sales.

Source 1, docs.together.ai, Billing and credits: Billing and creditsSource 2, Together AI, Provisioned throughput: Provisioned throughput

Details

Together AI is fully prepaid: you need a positive credit balance to use the platform, and access requires a minimum US$15 credit purchase. If your balance reaches zero, API access is suspended until you add credits. Customers with an active Scale or Enterprise contract continue to be billed under their existing contract terms. Serverless and dedicated endpoints are self-serve. Provisioned throughput is currently unavailable for self-service and is requested through sales.

Cost drivers

US$1.40 in, US$4.40 out

Per 1M tokens for GLM-5.3 on serverless. Dedicated GPUs bill by the minute, an H100 at US$5.49 an hour.

Source 9, docs.together.ai, Serverless overview: Serverless overviewSource 10, together.ai, GLM-5.3 model page: GLM-5.3 model pageSource 11, docs.together.ai, Dedicated endpoints pricing: Dedicated endpoints pricingSource 2, Together AI, Provisioned throughput: Provisioned throughputSource 12, docs.together.ai, Billing spend controls: Billing spend controls

Details

Serverless chat models bill per input and output token. GLM-5.3 costs US$1.40 per 1M input tokens and US$4.40 per 1M output tokens. Dedicated model inference bills by the minute for each running replica. An H100 80GB costs US$5.49 per hour. Provisioned throughput is priced at a flat US$0.05 per PTU per minute, on a one-month minimum term. Traffic above purchased PTU capacity may fall back to the shared serverless pool on a best-effort basis, billed at standard serverless rates.

Where it runs

US-only at 1.25x

us-api.together.ai keeps processing in the US for select models. EU needs a VPC or dedicated endpoint.

Source 13, Together AI, Privacy and security: Privacy and securitySource 14, docs.together.ai, Geo-specific inference: Geo-specific inferenceSource 9, docs.together.ai, Serverless overview: Serverless overviewSource 15, docs.together.ai, List placement profiles: List placement profilesSource 16, docs.together.ai, Dedicated endpoint configurations: Dedicated endpoint configurations

Details

For select serverless models, requests to us-api.together.ai keep inference processing in the United States at 1.25× the standard serverless rate. The guarantee covers only where those requests are processed and makes no promise about where data is stored, logged, or routed. For other data-residency, regulatory, or compliance requirements, such as GDPR, Together supports private networking and virtual private cloud (VPC) deployments, including in EU regions. Use a dedicated endpoint, or contact sales to discuss the right setup for your workload.

What to verify

SSO on Scale, Enterprise

The trust center lists SOC 2 and ISO 27001:2022. SCIM is listed under What's coming.

Source 18, docs.together.ai, Single sign-on: Single sign-onSource 17, trust.together.ai, Trust Center: Trust Center

Details

Together AI's trust center lists SOC 2 and ISO 27001:2022 under Compliance. SSO is available for Scale and Enterprise accounts. SCIM provisioning is listed under What's coming. Together's API lists an endpoint's audit and lifecycle events, newest first. That scope is per endpoint, and an organization-wide audit log of user or admin actions is not part of it.

Sourced from Together AI documentation