Category · Models & inference
GPU and serverless compute for AI workloads
Clouds that run your own models and jobs on GPUs by the second, as serverless functions, endpoints or rented machines. Compared on GPU types and price per hour, cold starts, scaling and what you have to package.
Best GPU and serverless compute for AI workloads, the shortlist231 head-to-head comparisons
Capability keys compute.gpu · compute.serverless · compute.endpoints · compute.batch · compute.containers · All tools
letme.dev/compute.gpu picks the top-graded tool in this list and says how to call it direct (how letme works).
The same listing from the live API. Graded results come first, then the official MCP registry when no graded-only filter is set.
https://www.anchorterminal.com/api/v1/search
Filters
| Compare | # | Tool | Category | Grade | Score | Agent rating | p95 | Context | Price / x402 | Auth | Where | Details |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 219 | Nebius AI CloudNebius · HTTP API | GPU compute | BB | 72.4 | 3.0 (4) | n/a | n/a | Pay per use | OAuth or key | Hosted and local | ||
|
Nebius AI Cloud rents NVIDIA GPU virtual machines and InfiniBand clusters, with managed Kubernetes, Slurm and Serverless AI jobs and endpoints for containers. Resources are managed through REST and gRPC APIs, a CLI, a Terraform provider and SDKs. Top strength OpenAPI 3.0.3 document at Top weakness The status page lists 27 incidents from 5 August to 31 October 2026, each limited to a region, service or group of nodes, among them about 21 hours of degradation in us-central1 |
||||||||||||
| 323 | Cloud Run GPUsGoogle Cloud · Platform | GPU compute | B | 69.6 | none | n/a | n/a | Pay per use | OAuth | Hosted | ||
|
Cloud Run GPUs attach one NVIDIA L4 or RTX PRO 6000 Blackwell GPU to a Cloud Run service, job or worker pool, so a container serves models serverlessly, billed per second and able to scale to zero. Top strength Scales to zero, bills per second, and instances with drivers start in about 5 seconds Top weakness One GPU per instance, L4 (24 GB) or RTX PRO 6000 (96 GB) only, so large models don't fit |
||||||||||||
| 405 | Hugging Face Inference EndpointsHugging Face · HTTP API | GPU compute | B | 67.9 | 3.0 (4) | n/a | n/a | $0.033 / vCPU-hr | OAuth or key | Hosted | ||
|
Managed Hugging Face service that deploys a Hub model as a dedicated, autoscaling HTTPS endpoint on AWS, Azure or Google Cloud, using vLLM, TGI, SGLang, llama.cpp, TEI or a custom container. Managed by REST API, Python client, CLI or MCP. Top strength Public OpenAPI 3.1 documents for the management API (46 operations) and the catalogue API (3), plus llms.txt and a Markdown twin of every docs page Top weakness No free tier. The docs require a payment method and credits, and replicas are billed while initialising as well as running |
||||||||||||
| 436 | Modal SandboxesModal · SDK + MCP | Sandboxes | B | 67.3 | 3.3 (8) | n/a | n/a | $0.071 / vCPU-hr | API key | Local | ||
|
Modal's sandboxed compute environments for running code, with SDK access, GPU support and filesystem snapshots. Top strength GPU sandboxes at the same per-second rates as the rest of Modal Top weakness No REST API, and the JavaScript and Go SDKs are beta |
||||||||||||
| 453 | Replicate DeploymentsReplicate (Cloudflare) · HTTP API | GPU compute | B | 67 | 3.0 (2) | n/a | n/a | Pay per use | API key | Hosted and local | ||
|
Replicate's service for deploying and running custom models. Top strength OpenAPI file, llms.txt and an MCP server with a two-tool code mode Top weakness Private instances bill set-up and idle time, H100 at $5.49 an hour |
||||||||||||
| 532 | BasetenBaseten · HTTP API | GPU compute | B | 65.2 | 3.5 (2) | n/a | n/a | Pay per use | API key | Hosted | ||
|
Dedicated model deployments packaged with the open-source Truss framework and served behind a per-model HTTPS endpoint, with autoscaling from zero replicas, async inference, a management API and per-minute GPU billing from T4 to B200. Top strength Team API keys scoped to inference-only, metrics-only or a single environment or model, plus a Viewer role since 1 September 2026 Top weakness H100 at $6.50 and A100 at $4.00 an hour, and start-up and idle replica time are billed |
||||||||||||
| 582 | Crusoe CloudCrusoe · HTTP API | GPU compute | B | 64.3 | 3.3 (4) | n/a | n/a | $0.04 / vCPU-hr | API key | Hosted and local | ||
|
Crusoe Cloud rents NVIDIA and AMD GPU virtual machines with managed Kubernetes and Slurm, and runs hosted model inference, dedicated deployments and LoRA fine-tuning. Resources are managed through a REST API, a CLI, a Terraform provider and a Go client. Top strength Requests are signed with HMAC-SHA256 over the path, query, verb and timestamp, so the secret key never travels with a request. Top weakness No idempotency key or safe-retry guidance was found for create and delete calls, which return asynchronous operations to poll. |
||||||||||||
| 599 | ModalModal · Platform | GPU compute | B | 63.9 | 4.0 (2) | n/a | n/a | $250 / mo | API key | Local | ||
|
Serverless functions, web endpoints, servers and GPU jobs from a Python decorator, with JavaScript and Go SDKs. Top strength Scale to zero by default, per-second billing and about one-second container boots Top weakness No REST API or OpenAPI spec for deploying or invoking Functions |
||||||||||||
| 691 | CoreWeaveCoreWeave, Inc. · HTTP API | GPU compute | C | 61.8 | 2.5 (4) | n/a | n/a | Pay per use | Token | Hosted | ||
|
CoreWeave is a GPU cloud that rents NVIDIA GPU nodes through a managed Kubernetes service (CKS), with REST and gRPC platform APIs, a Terraform provider and a hosted MCP server for observability. Top strength On-demand and spot prices per instance-hour are public, for example HGX H100 at $49.24 on demand and $19.71 spot for 8 GPUs Top weakness No self-serve route. CoreWeave's sales team approves an organisation and emails the activation link before a token can be created |
||||||||||||
| 800 | Together GPU ClustersTogether AI · HTTP API | GPU compute | C | 59.4 | none | n/a | n/a | Pay per use | API key | Hosted | ||
|
Together GPU Clusters rents multi-node NVIDIA H100, H200, B200 and B300 clusters with InfiniBand and shared storage, run as Kubernetes or Slurm, by the GPU-hour. Clusters are created and scaled from the console, the tg CLI, SDKs or REST API. Top strength Public OpenAPI 3.1, llms.txt and Markdown docs, with an automation guide for CI pipelines Top weakness Provisioning takes 10 minutes to an hour, and a cluster can't be paused or scaled to zero |
||||||||||||
| 864 | Vast.aiVast.ai Inc. · HTTP API | GPU compute | C | 57.6 | 3.0 (4) | n/a | n/a | Pay per use | API key | Hosted | ||
|
Vast.ai is a marketplace for renting GPUs by the second from independent hosts and data centres, as Docker instances, virtual machines or autoscaling serverless endpoints. Agents use the `vastai` CLI, a Python SDK or a REST API. Top strength API keys take 11 permission categories and per-endpoint constraints on resource IDs, and can be reset or deleted at once Top weakness No SLA. The terms say availability is not guaranteed and the service can change without notice |
||||||||||||
| 873 | BeamBeam · Platform | GPU compute | C | 57.4 | 3.0 (2) | n/a | n/a | $0.045 / vCPU-hr | API key | Hosted | ||
|
Beam runs persistent VMs, sandboxes, serverless GPU endpoints, task queues and functions for agents and Python code, billed by the millisecond (VMs by the second), on the open-source beta9 runtime. Top strength Per-millisecond billing with cold starts and image pulls free, RTX 4090 at $0.69 an hour serverless Top weakness No free plan; the Developer plan costs $20 a month, with $20 of credits |
||||||||||||
| 894 | RunpodRunpod · HTTP API | GPU compute | C | 56.9 | 3.0 (2) | n/a | n/a | Pay per use | OAuth or key | Hosted and local | ||
|
Serverless GPU endpoints, queue-based or load-balanced, and rented GPU Pods, billed per second from prepaid credit. Top strength Per-second billing across more than a dozen serverless GPU classes, H100 at $4.79 and A100 80 GB at $2.72 an hour Top weakness Data-centre outages of 6 to 24 hours in each of July, August and September 2026, per the 1 October check |
||||||||||||
| 920 | NorthflankNorthflank · Platform | GPU compute | C | 56.5 | 3.5 (2) | n/a | n/a | $0.0167 / vCPU-hr | Token | Hosted | ||
|
Platform for deploying services, jobs and databases from Git or container images, in managed or customer-owned infrastructure. Supports GPU workloads and sandboxes. Top strength OpenAPI 3.0 with over 100 paths, enums and Top weakness A security incident opened on 30 October 2026, still open on 2 November, told customers to assume all environment variables were decrypted |
||||||||||||
| 991 | VerdaVerda · HTTP API | GPU compute | C | 54.5 | 3.3 (4) | n/a | n/a | Pay per use | OAuth | Hosted and local | ||
|
Verda, formerly DataCrunch, is a Finnish GPU cloud renting GPU and CPU instances, clusters, serverless containers and storage. Agents use its REST API, Python and Go SDKs, or a CLI with a built-in MCP server in beta. Top strength
Top weakness No free tier. Accounts are prepaid, and instances are discontinued and volumes deleted when the balance reaches zero |
||||||||||||
| 994 | CerebriumCerebrium Inc. · Platform | GPU compute | C | 54.4 | none | n/a | n/a | $0.0236 / vCPU-hr | API key | Hosted | ||
|
Cerebrium is a serverless platform for running your own models and code on GPUs and CPUs. A CLI packages code into containers served as REST, streaming and WebSocket endpoints, managed through a REST API. Top strength Per-second prices for ten GPU types published without a login, from T4 at $0.59 an hour to B200 at $6.01 Top weakness
|
||||||||||||
| 1007 | Lambda CloudLambda · HTTP API | GPU compute | C | 54.1 | 3.0 (2) | n/a | n/a | Pay per use | API key | Hosted | ||
|
On-demand GPU virtual machines and clusters, with an API for provisioning compute and persistent storage. Top strength H100 SXM at $3.99 and B200 at $6.69 an hour with per-minute billing Top weakness No scale to zero, autoscaling or endpoints; an idle VM bills until terminated |
||||||||||||
| 1062 | Thunder ComputeThunder Compute · HTTP API | GPU compute | D | 52.5 | none | n/a | n/a | $0.04 / vCPU-hr | OAuth or key | Hosted | ||
|
Thunder Compute is a US GPU cloud renting RTX A6000, L40, A100 and H100 instances by the minute and GPU sandboxes by the second. Agents use its REST API, a hosted MCP server with OAuth, or the `tnr` CLI. Top strength Hosted MCP server at Top weakness No rate limits, 429 guidance or SLA were found in the reviewed documentation, and the terms disclaim availability |
||||||||||||
| 1196 | KoyebKoyeb · Platform | GPU compute | D | 47.8 | 3.5 (2) | n/a | n/a | $29 / mo | Token | Hosted | ||
|
Serverless platform for deploying applications and containers on CPU or GPU instances, with autoscaling and an API. Top strength H100 at $2.50, H200 at $3.00 and B200 at $5.50 an hour, billed per second Top weakness Public changelog silent since 27 February 2026, so API changes surface only through CLI and SDK releases |
||||||||||||
| 1299 | Massed ComputeMassed Compute · HTTP API | GPU compute | E | 43.3 | 3.3 (4) | n/a | n/a | Pay per use | OAuth or key | Hosted | ||
|
Massed Compute is a US GPU cloud that rents NVIDIA GPU virtual machines by the hour from hardware it owns, with bare metal and clusters by quote. Agents use its REST API or a hosted MCP server. Top strength API tokens and OAuth grants come in read-only and full-access tiers, and tokens are revoked one at a time in the console Top weakness No status page, incident history or security.txt was found on any of the vendor's hosts |
||||||||||||
| 1331 | GPUsMarketTyga.Cloud Ltd · HTTP API | GPU compute | E | 41.5 | none | n/a | n/a | $0.50 / tx | API key | Hosted and local | ||
|
GPUsMarket is a peer-to-peer GPU marketplace from Tyga.Cloud Ltd. Agents rent hosts' NVIDIA GPUs and Macs, or on-demand cloud GPUs, as Ollama and OpenAI-compatible endpoints billed per second, through a REST API, CLI and MCP server. Top strength Sign-up and key issue in one API call, and search and GPU prices need no key Top weakness Peer rentals send prompts and outputs through individuals' hardware, with only the terms as protection |
||||||||||||
| 1360 | HyperbolicHyperbolic Labs, Inc. · HTTP API | GPU compute | E | 39.7 | none | n/a | n/a | Pay per use | API key | Hosted | ||
|
Hyperbolic is a GPU cloud from Hyperbolic Labs that rents H100, H200 and B200 virtual machines and bare-metal nodes by the hour, with reserved terms and private clusters. Agents use its REST API with an API key. Top strength Deprecated paths answer with a Top weakness API keys carry no scopes or expiry in the reviewed documentation, and the same key can call |
||||||||||||
Nothing matches these filters. .
p95 latency and context cost come from our probes, which haven't run yet, so those columns start hidden. Grades run from AA to F, and agent-ready means BB or better. Filters, sorting and export run in your browser; the table is complete without JavaScript.
Indexed, not reviewed (9)
Listings sorted into this category from public catalogues (the official MCP registry, APIs.guru, the x402 Bazaar and OpenRouter), with facts and our own checks but no score, grade or rank. How the index works.
| Listing | Kind | What it does | Why it's here |
|---|---|---|---|
| AI Compute Radar aicomputeradar.dev | MCP server | Which open models fit your GPU or Mac, measured. Model momentum, GPU rental prices, weekly pick. | vendor's own |
| AI Video Enhancer Studio contenta-software.com | MCP server | Upscale, stabilize, denoise and raise the frame rate of videos locally on your NVIDIA GPU | vendor's own |
| AmpleRun GPU rentals amplerun.com | MCP server | Find, price, rent and stop GPUs on AmpleRun. Paid in USDC or USDT on Base. Flat 5% fee. | vendor's own |
| matrix-compute-mcp matrix-compute.com | MCP server | P2P GPU matrix multiply, exact cosine search, and PCA/SVD verified via Freivalds' algorithm. | vendor's own |
| Nodegrove VRAM: can I run it? nodegrove.io | MCP server | Can this LLM run on my GPU? VRAM, speed ceiling and what fits instead, for any model and GPU. | vendor's own |
| prismnetwork.tech MCP server prismnetwork.tech | MCP server | Rent real NVIDIA GPUs from your agent. Browse with no wallet; pay per second onchain. | vendor's own |
| rent.dev rent.dev | MCP server | Linux sandboxes billed by the second from a prepaid balance, and GPU pods on your own RunPod key. | vendor's own |
| singularis getsingularis.com | MCP server | Unlimited computation for agents. Results depend on your equipment; GPU optional. | vendor's own |
| VideoRecompress Studio contenta-software.com | MCP server | Shrink video files on your Windows PC with H.265/AV1 and GPU encoding, a whole folder at a time | vendor's own |
How we test this category
The same model deployed as an endpoint on each platform, called cold and warm, then scaled to zero. We time cold starts, check the scaling and add up the cost per GPU-hour.
How the ranking works
Every tool is scored 0 to 100 across 9 weighted categories, minus deductions for negative events, and given a grade from AA to F. Reliability, latency and context cost are measured by probes and by traffic through letme. Schema quality, security and payments are assessed against a published checklist. Task success comes from a fixed task suite. Payments, including x402, are one category of the nine. Read the methodology.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.