Together AI
US AI cloud centred on per-token open-model inference that also sells self-serve NVIDIA HGX GPU clusters (8 GPUs minimum) on preemptible, on-demand and prepaid reserved terms, plus per-minute dedicated inference endpoints.
Trust Score v0.1- Company & track record73
- Infrastructure control100
- Pricing & billing transparency85
- Reliability signals55
- Security & compliance100
- Customer signal—
Scored Sep 29, 2026 from the provider's own pages. Dimensions not yet measured are excluded from the denominator. Criteria ↓
Trust Score · 5 of 6 dimensions measuredThe score, criterion by criterion
Each point comes from something the provider publishes, with the page it was read on and the date. Anything not disclosed earns zero.
| Founding year published | not disclosedtogether.ai · Sep 29, 2026 | 0 / 4 |
| Headquarters published | Together Computer, Inc.together.ai · Sep 29, 2026 | 3 / 3 |
| Legal entity named | Together Computer, Inc.together.ai · Sep 29, 2026 | 4 / 4 |
| Funding or ownership disclosed | Private. Announced an 800M USD Series C on 1 Jul 2026 (investors incl. Aramco Ventures, NVIDIA, Vista Equity, General Catalyst) plus commitments for over 500 MW of compute capacity financed by the new investors.together.ai · Sep 29, 2026 | 4 / 4 |
| Capacity model published | operates or rents dedicated capacitytogether.ai · Sep 29, 2026 | 10 / 10 |
| Regions published | 2 regions or countriestogether.ai · Sep 29, 2026 | 5 / 5 |
| Public prices without an account | yes (read on the pricing page)together.ai · Sep 29, 2026 | 5 / 5 |
| Billing granularity published | mixedtogether.ai · Sep 29, 2026 | 4 / 4 |
| Egress policy published | includedtogether.ai · Sep 29, 2026 | 4 / 4 |
| Stopped-instance billing published | GPU clusters: on-demand clusters bill hourly until deleted; the billing FAQ states there is no pause function (another section of the same page says on-demand clusters can be stopped/started). Shared volumes keep billing at 0.16 USD/GiB-month after cluster deletion. Dedicated inference: a stopped deployment or zero replicas costs nothing.together.ai · Sep 29, 2026 | 4 / 4 |
| Public IPv4 policy published | not disclosedtogether.ai · Sep 29, 2026 | 0 / 3 |
| Status page | https://status.together.ai/together.ai · Sep 29, 2026 | 6 / 6 |
| SLA published | not disclosedtogether.ai · Sep 29, 2026 | 0 / 6 |
| SLA with a numeric uptime commitment | not disclosedtogether.ai · Sep 29, 2026 | 0 / 3 |
| Uptime history published | Status page shows 100% 90-day uptime for Website and Playground only; no GPU cluster component listed.together.ai · Sep 29, 2026 | 5 / 5 |
| Observed availability (30 days) | not measured yet | — |
| SOC 2 | Type IItogether.ai · Sep 29, 2026 | 8 / 8 |
| ISO 27001 | ISO 27001:2022 certified (announced 29 May 2026)together.ai · Sep 29, 2026 | 4 / 4 |
| HIPAA or GDPR DPA | DPA available via Trust Centertogether.ai · Sep 29, 2026 | 3 / 3 |
| Aggregated public reviews | not collected yet | — |
Prices by GPU · 3 GPU types with a current offerWhat Together AI charges, against the market
Cheapest current on-demand true cost here, next to the market median for the same GPU. Negative means cheaper than the market.
| GPU | Cheapest here | Market median | vs market | Current offers | Modes | |
|---|---|---|---|---|---|---|
| H100 | $3.99 | $3.17 | +26% | 2 | on-demandspot | H100 prices |
| H200 | $5.99 | $4.76 | +26% | 2 | on-demandspot | H200 prices |
| B200 | $8.19 | $7.50 | +9% | 2 | on-demandspot | B200 prices |
All offers · 40 readings on fileEvery Together AI offer, with its receipt
Current offers (last 45 days) ranked by true cost, all rental modes, each with its read date. Older readings stay listed below, in grey.
| GPU | Configuration | Mode | Region | Stock | List | True cost | Read | |
|---|---|---|---|---|---|---|---|---|
| H100SXM | 1× H100 SXM | Spot | — | Stock unknown | $1.99 | $1.99 | 21:35 UTCfresh | Receipt |
| H200 | 1× H200 | Spot | — | Stock unknown | $2.99 | $2.99 | 21:35 UTCfresh | Receipt |
| H100SXM | 1× H100 SXM | On-demand | — | Stock unknown | $3.99 | $3.99 | 21:35 UTCfresh | Receipt |
| B200 | 1× B200 | Spot | — | Stock unknown | $4.09 | $4.09 | 21:35 UTCfresh | Receipt |
| H200 | 1× H200 | On-demand | — | Stock unknown | $5.99 | $5.99 | 21:35 UTCfresh | Receipt |
| B200 | 1× B200 | On-demand | — | Stock unknown | $8.19 | $8.19 | 21:35 UTCfresh | Receipt |
True cost & billing · read Sep 29, 2026What the hourly price does not say
Billing granularity, minimums, what a stopped machine costs, traffic, addresses, deposits and identity checks, as published by the provider.
Billing
- Billing granularity
- mixed
- Minimum charge
- Platform access requires a minimum 5 USD credit purchase. GPU clusters are created in multiples of 8 GPUs. Reservations are charged upfront for the full term and are non-refundable. Shared volume minimum 1 TiB.
- Stopped instance
- GPU clusters: on-demand clusters bill hourly until deleted; the billing FAQ states there is no pause function (another section of the same page says on-demand clusters can be stopped/started). Shared volumes keep billing at 0.16 USD/GiB-month after cluster deletion. Dedicated inference: a stopped deployment or zero replicas costs nothing.
- Egress
- included
- Public IPv4
- not published
- Billing model
- prepaid credits (self-serve is fully prepaid; API access suspended at zero balance); Scale/Enterprise contracts billed under contract terms
- Minimum deposit
- 5 USD minimum credit purchase
- Free credits
- No free trial offered.
Account
- Payment
- card
- Sign-up
- Google or GitHub OAuth account (no email/password option), credit or debit card (prepaid cards refused; one valid card must always be on file), 5 USD minimum credit purchase
- Identity check
- card verification
- Support
- support portal tickets, email, discord, shared slack (organisations that have one)
- Enterprise support
- not published
- SLA
- not published
- Status page
- status.together.ai ↗
- Published uptime
- Status page shows 100% 90-day uptime for Website and Playground only; no GPU cluster component listed.
Infrastructure
- Regions
2 named regions in 5 countries
us-central-8 (United States) · us-central-2 (United States)- Countries
- United States, France, Netherlands, Sweden, Romania
- Platform
- API, CLI, serverless, SSH, custom Docker images, persistent volumes, spot or interruptibleSDK: python, typescriptTerraform: officialKubernetes: managedInterconnect: InfiniBand (speed not published)Clusters: Self-serve Kubernetes or Slurm-on-Kubernetes clusters: 8 to 256 GPUs on demand, up to 4,000+ GPUs reserved; SkyPilot integration; GB200/GB300 NVL72 via sales.
- Storage
- network · 0.16 USD/GiB-month · Shared filesystem (WEKA or VAST) attached to GPU clusters; persistent, resizable upward self-serve (downsizing via account team), minimum 1 TiB; volumes of reserved clusters move to on-demand pricing at expiry.ephemeral · not itemised · Local NVMe on each node; data can be lost on node migration, maintenance or failure.
- Migration out
- Shared volumes persist independently of clusters and can be copied out with kubectl cp or with pods pushing to S3; there are no egress fees on GPU clusters. Workloads are standard Kubernetes or Slurm jobs with customer container images, so they move to any Kubernetes or Slurm environment. Dedicated endpoints use an OpenAI-compatible API.
Security & compliance
- SOC 2
- Type II
- ISO 27001
- ISO 27001:2022 certified (announced 29 May 2026)
- HIPAA
- not published
- GDPR / DPA
- DPA available via Trust Center
- Other
- NVIDIA Cloud Partner; private HackerOne vulnerability program; zero data retention option for inference
- Legal entity
- Together Computer, Inc.
- Funding
- Private. Announced an 800M USD Series C on 1 Jul 2026 (investors incl. Aramco Ventures, NVIDIA, Vista Equity, General Catalyst) plus commitments for over 500 MW of compute capacity financed by the new investors.
Assessment · documents read Sep 29, 2026Together AI, in short
What stood out when we read the documents: strengths, weaknesses, who it suits, and what we could not find.
Strengths
- Public per-GPU-hour prices for H100, H200, B200 and B300 across preemptible, on-demand and three reserved tiers.
- Managed Kubernetes and Slurm-on-Kubernetes clusters with InfiniBand, created self-serve in minutes, up to 256 GPUs on demand and 4,000+ reserved.
- No ingress or egress fees on GPU clusters.
- One account covers per-token serverless inference, per-minute dedicated endpoints, fine-tuning and GPU clusters.
- SOC 2 Type II report and ISO 27001:2022 certificate listed on the Trust Center.
- Official Terraform provider, CLI (tg), Python/TypeScript SDKs, REST API and SkyPilot support.
Weaknesses
- Minimum unit is 8 GPUs; no single-GPU VM for general compute.
- On-demand H100 at 3.99 USD/GPU-h, H200 at 5.99 and B200 at 8.19.
- Reservations are paid upfront and non-refundable; reserved clusters cannot be paused.
- Card or US ACH only; prepaid and virtual cards often refused; no crypto.
- Sign-up only through Google or GitHub OAuth.
- No SLA in the public terms (only via order form); GPU clusters are labelled beta in the CLI docs.
- Shared storage minimum 1 TiB and cannot be downsized self-serve.
Ideal for
- Teams training or fine-tuning on 8 to 256 Hopper or Blackwell GPUs who want managed Kubernetes or Slurm
- Companies already on Together serverless inference that want dedicated endpoints or training capacity on the same bill
- Buyers who need SOC 2 Type II and ISO 27001 documentation
Not published
- Year founded not stated on the official pages read (the Jul 2026 Series C post says the company was started 'four years ago').
- Per-node vCPU, RAM and local NVMe for cluster GPU types not published on the pricing or docs pages read.
- Full region list and per-region GPU availability are only returned by the authenticated regions API
- public docs show example regions us-central-8 and us-central-2. France, Netherlands, Sweden and Romania are named as European capacity locations, not as self-serve regions.
- Prices for RTX_6000_PCI, L40_PCIE and H100_SXM_INF cluster GPU types (listed in the API enum) not published
- the exact RTX 6000 variant (Ada or PRO) is not stated.
- Reserved 181+ days, reserved B300, dedicated inference H200/B300/GB300, and GB200/GB300 NVL72 cluster prices: contact sales.
- IPv4 / public IP policy not published.
- HIPAA: the GPU clusters page mentions choosing regions for HIPAA requirements, but no HIPAA attestation or BAA is published.
- No public SLA
- the terms say no uptime guarantee unless agreed in an order form.
- Snapshots, Jupyter and templates are not documented for GPU clusters.
Observations
Pricing page is public, no account needed; it opens with per-token serverless model prices and places GPU Clusters after Provisioned Throughput and Dedicated Inference.
GPU Clusters table states 'All prices are per GPU per hour' with columns Preemptible compute, On-demand and Reserved (7-30, 31-90, 91-180 days, 181+ days 'Contact us').
GB200 NVL72 and GB300 NVL72 rows show only dashes and 'Contact us'; H200, B300, GB200 and GB300 dedicated inference rows show 'Contact us'.
Dedicated inference H100 carries a PROMO badge: 5.49 struck through, 3.99, 'PROMOTION VALID UNTIL 09/30/26'.
Site-wide banner: 'On-demand B200s now available on Together GPU Clusters'.
Docs describe reserved terms of 1-90 days while the pricing page shows tiers to 181+ days and the GPU clusters page says reserved 'up to 6 months'.
Prices in USD only, taxes excluded per the terms of service; no vCPU/RAM specs next to cluster prices.
GPU clusters page calls to action are 'Create cluster' and 'Reserve capacity'; the header shows 'Contact sales' and 'Sign in'.
Pages read
- together.ai/pricing — GPU cluster, dedicated inference and storage prices
- together.ai/gpu-clusters — Scale limits, VRAM, reserved terms, capacity countries, SOC 2 and ISO badges
- docs.together.ai/docs/gpu-clusters-overview — Capacity types, preemptible, storage tiers, InfiniBand, hardware
- docs.together.ai/docs/gpu-clusters-billing — Billing, non-refundable reservations, no egress fees, pause policy, credit exhaustion
- docs.together.ai/docs/preemptible-compute — Preemption mechanics and metering
- docs.together.ai/docs/cluster-storage — Persistent vs ephemeral storage
- docs.together.ai/docs/gpu-clusters-quickstart — Reserved 1-90 days, 1 TiB minimum volume, SSH access
- docs.together.ai/docs/dedicated-endpoints/pricing — Per-minute replica billing, hardware price table
- docs.together.ai/docs/billing-credits — Prepaid model, 5 USD minimum, no free trial
- docs.together.ai/docs/billing-payment-methods — Cards, ACH limits, card-on-file rule
- docs.together.ai/docs/billing-troubleshooting — Prepaid/virtual cards refused, 0 USD authorization hold
- docs.together.ai/docs/account-management — Google/GitHub OAuth sign-up
- docs.together.ai/reference/clusters-create — GPU type enum, multiples of 8, billing types, Terraform mention
- docs.together.ai/reference/cli/clusters — CLI, beta status, example region
- docs.together.ai/docs/gpu-clusters-api — CLI, REST, SkyPilot
- github.com/togethercomputer/terraform-provider-together — Official Terraform provider
- docs.together.ai/docs/dedicated-container-inference — Custom Docker inference, access via sales
- docs.together.ai/docs/support — Support channels
- docs.together.ai/docs/privacy-and-security — EU residency via dedicated/VPC deployments, HackerOne
- trust.together.ai/ — SOC 2 Type 2, ISO 27001:2022, DPA, subprocessors (Stripe payments
- status.together.ai/ — Status page
- together.ai/terms-of-service — Legal entity, address, taxes excluded, no uptime guarantee
- together.ai/blog/announcing-our-series-c — Funding
- together.ai/about-us — Founders and leadership
Questions about Together AI
What GPUs does Together AI offer?
Together AI has 4 GPU types with a fresh offer today, for example H100 from $3.99, H200 from $5.99, B200 from $8.19 per GPU-hour on-demand. The full list with prices and market medians is on this page.
How does Together AI bill?
Billing granularity: mixed. Minimum charge: Platform access requires a minimum 5 USD credit purchase. GPU clusters are created in multiples of 8 GPUs. Reservations are charged upfront for the full term and are non-refundable. Shared volume minimum 1 TiB. Stopped instance: GPU clusters: on-demand clusters bill hourly until deleted; the billing FAQ states there is no pause function (another section of the same page says on-demand clusters can be stopped/started). Shared volumes keep billing at 0.16 USD/GiB-month after cluster deletion. Dedicated inference: a stopped deployment or zero replicas costs nothing.
How trustworthy is Together AI?
Its Trust Score is 81/100 (v0.1), read from its own documents. Strong on infrastructure control, security & compliance. Weaker on reliability signals. The score measures transparency and published guarantees, not customer satisfaction; every criterion and its source is listed above.
Where are Together AI's data centres?
2 named regions in 5 countries: United States, France, Netherlands, Sweden, Romania.
How do you pay Together AI, and are there credits?
Payment by card. Free credits: No free trial offered.. Identity check: card verification.