{ "data": { "a": { "slug": "cloud-run-gpus", "name": "Cloud Run GPUs", "vendor": "Google Cloud", "vendorUrl": "https://cloud.google.com/run", "kind": "platform", "category": "gpu-compute", "summary": "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.", "url": "https://www.anchorterminal.com/tools/cloud-run-gpus", "markdownUrl": "https://www.anchorterminal.com/tools/cloud-run-gpus.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/cloud-run-gpus.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/cloud-run-gpus.json", "dns": "cloud-run-gpus.grade.anchorterminal.com", "license": "Proprietary service under the Google Cloud Platform terms. Client libraries are Apache-2.0", "transports": [], "remoteUrl": "https://run.googleapis.com", "packages": [ { "registry": "pypi", "name": "google-cloud-run" }, { "registry": "npm", "name": "@google-cloud/run" } ], "auth": "oauth", "authNotes": "Deploys go through the Cloud Run Admin API with OAuth 2.0 bearer tokens from a service account or Application Default Credentials, under IAM roles such as roles/run.developer and roles/run.invoker. A deployed service can require IAM authentication or allow public calls. A project with billing enabled is set up in the console first.", "pricing": "usage", "pricingNotes": "Per second, instance-based billing only. L4 $0.0001867 a second (about $0.67 an hour) without zonal redundancy, $0.0002909 with; RTX PRO 6000 $0.00036522 (about $1.31 an hour) and $0.00056913. CPU and memory extra, with a minimum of 4 vCPU and 16 GiB (L4) or 20 vCPU and 80 GiB (RTX PRO 6000).", "priceSummary": "Pay per use", "where": "hosted", "x402": { "level": "no", "evidence": "No x402, MPP or L402 on the Cloud Run pricing, GPU or quotas pages (checked 2026-11-06). Billing runs through a Google Cloud billing account.", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-11-06" }, "docsUrl": "https://docs.cloud.google.com/run/docs/configuring/services/gpu", "openapi": "https://run.googleapis.com/$discovery/rest?version=v2", "capabilities": [ "compute.serverless", "compute.gpu", "compute.endpoints", "compute.containers", "compute.batch" ], "tags": [ "hosted", "official", "usage", "per-second", "scale-to-zero", "sla", "status-page", "discovery-doc", "l4", "rtx-pro-6000" ], "lastRelease": "2026-10-30", "graded": true, "anchor": { "graded": true, "score": 68.4, "grade": "B", "agentReady": false, "rank": 289, "ranked": true, "rankOf": 1510, "categoryRank": 2, "methodology": "0.7", "run": "2026-10-01", "scores": { "ergonomics": 66, "maintenance": 80, "payments": 20, "reliability": 70, "schema": 81, "security": 78, "transparency": 84 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "medium", "date": "2026-11-06" }, "negative": 0, "verdict": "Serverless GPUs on Cloud Run, an L4 or RTX PRO 6000 per instance, billed per second, scaling to zero and starting in about 5 seconds before the model loads, with a published SLA and IAM. One GPU per instance, 3 GPUs of default quota a region and instance-based billing limit it to small and mid-size models.", "bestFor": "Serving small and mid-size open models, or batch jobs, from a container that should scale to zero.", "strengths": [ "Scales to zero, bills per second, and instances with drivers start in about 5 seconds", "SLA of 99.95% with zonal redundancy and 99.5% without, plus Google's status feed", "IAM roles for deploying and invoking, Cloud Audit Logs and services that can require authentication", "Per-second prices public, L4 from about $0.67 a GPU-hour without zonal redundancy", "The same GPU option on services, jobs (up to 12 hours) and worker pools" ], "weaknesses": [ "One GPU per instance, L4 (24 GB) or RTX PRO 6000 (96 GB) only, so large models don't fit", "Default quota is 3 GPUs a region; more needs a quota request", "Model load time comes on top of the 5-second start, and minimum instances bill while idle", "No free GPU tier, and the trial credit needs a payment method", "Google's APIs accept OAuth tokens in a URL query parameter" ], "agentNotes": [ "Pass `--gpu 1 --gpu-type nvidia-l4` (or nvidia-rtx-pro-6000) with at least 4 vCPU and 16 GiB, and `--no-cpu-throttling` for instance-based billing", "Load model weights from Cloud Storage at startup in parallel; bake them into the image only below about 10 GB", "Turn zonal redundancy off for cheaper, more readily granted quota when failover isn't required", "Set minimum instances to 0 unless cold starts matter, since minimum instances bill at the full rate", "Send OAuth tokens in the `Authorization` header, never as `access_token` in the URL" ], "metrics": { "kind": "remote", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "medium", "grade": "B", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 68.4 } ], "editorialScores": { "ergonomics": 66, "maintenance": 80, "payments": 20, "reliability": 70, "schema": 81, "security": 78, "transparency": 69 }, "provenanceScore": 99, "checked": { "categories": { "ergonomics": "2026-11-06", "maintenance": "2026-11-06", "payments": "2026-11-06", "reliability": "2026-11-06", "schema": "2026-11-06", "security": "2026-11-06", "transparency": "2026-11-06" }, "last": "2026-11-06", "oldest": "2026-11-06", "stale": false, "staleAfterDays": 90 } }, "connect": { "headless": { "command": "gcloud run deploy SERVICE --image IMAGE --gpu 1 --gpu-type nvidia-l4 --cpu 4 --memory 16Gi --no-cpu-throttling --region europe-west1" } }, "letme": { "capability": "https://letme.dev/compute.serverless", "tool": "https://letme.dev/cloud-run-gpus" }, "sameCompany": [ "gemini-api", "vertex-ai", "gemini-embedding", "vertex-ai-tuning", "google-model-armor", "google-imagen", "google-veo", "google-lyria", "google-speech-to-text", "google-cloud-tts", "gemini-live", "google-adk", "google-memory-bank", "google-secret-manager", "google-cloud-document-ai", "google-weather-api", "chrome-devtools-mcp", "google-maps-platform", "google-cloud-translation", "google-calendar-api", "firebase-cloud-messaging", "google-cloud-storage", "google-drive-api", "gemini-enterprise", "gemini-cli", "google-search-console", "google-ads-api", "google-forms", "google-sheets-api", "gmail-api", "google-people-api" ], "area": "models", "unitPrices": [ { "item": "NVIDIA L4, no zonal redundancy", "unit": "gpu-hour", "usd": 0.67, "note": "$0.0001867 a second, tier 1 regions; CPU and memory extra" }, { "item": "NVIDIA L4, zonal redundancy", "unit": "gpu-hour", "usd": 1.05, "note": "$0.0002909 a second" }, { "item": "NVIDIA RTX PRO 6000, no zonal redundancy", "unit": "gpu-hour", "usd": 1.31, "note": "$0.00036522 a second" }, { "item": "NVIDIA RTX PRO 6000, zonal redundancy", "unit": "gpu-hour", "usd": 2.05, "note": "$0.00056913 a second" } ], "provenance": { "legalEntity": "Google LLC", "domain": "google.com", "domainRegistered": "1997-09-15", "domainNote": "The endpoints are on googleapis.com, Google's API domain. Docs are on docs.cloud.google.com.", "endpointOnVendorDomain": true, "terms": "https://cloud.google.com/terms", "privacy": "https://cloud.google.com/terms/cloud-privacy-notice", "statusPage": "https://status.cloud.google.com", "changelog": "https://docs.cloud.google.com/run/docs/release-notes", "securityTxt": "valid", "checked": "2026-11-06", "notes": [ "Legal entity, terms, privacy notice and security.txt as on the Google Cloud Document AI listing, checked 1 November 2026, and not re-read for this listing.", "docs.cloud.google.com/robots.txt and cloud.google.com/robots.txt allow the Cloud Run pages read; status.cloud.google.com/robots.txt allows everything.", "The Cloud Run SLA lists GPU services separately, 99.95% with zonal redundancy and 99.5% without (https://cloud.google.com/run/sla)." ], "score": 99 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/cloud-run-gpus.json", "live": { "slug": "cloud-run-gpus", "probe": { "target": "https://run.googleapis.com", "method": "get", "lastAt": "2026-11-07T22:55:51.650380142Z", "lastOk": true, "lastStatus": 404, "lastMs": 68, "authRequired": false, "uptime24h": 100, "uptime30d": 100, "p50ms24h": 37, "p95ms24h": 158, "samples24h": 244, "samples30d": 246, "days": [ { "date": "2026-11-06", "probes": 13, "ok": 13 }, { "date": "2026-11-07", "probes": 233, "ok": 233 } ] }, "vendorStatus": { "page": "https://status.cloud.google.com", "indicator": "unknown", "summary": "no machine-readable status found", "checkedAt": "2026-11-07T04:47:16.629467571Z" }, "versions": [ { "registry": "npm", "name": "@google-cloud/run", "version": "4.2.0", "seenAt": "2026-11-06T23:15:29.275446701Z" }, { "registry": "pypi", "name": "google-cloud-run", "version": "0.16.2", "released": "2026-10-01", "seenAt": "2026-11-06T23:15:27.277547645Z" } ], "npmWeekly": 1706675, "pypiWeekly": 3039289, "securityTxt": { "url": "https://google.com/.well-known/security.txt", "state": "valid", "expires": "2030-04-01T00:00:00z", "checkedAt": "2026-11-07T16:01:22.068349931Z" }, "pages": [ { "url": "https://docs.cloud.google.com/run/docs/release-notes", "kind": "changelog", "status": 200, "checkedAt": "2026-11-07T14:05:47.69894113Z", "changedAt": "0001-01-01T00:00:00Z", "fingerprint": "924ee0205e11" }, { "url": "https://cloud.google.com/terms/cloud-privacy-notice", "kind": "privacy", "status": 200, "checkedAt": "2026-11-07T14:01:29.495526727Z", "changedAt": "2026-11-03T07:43:24.319773868Z", "fingerprint": "b62369d64ac6" }, { "url": "https://cloud.google.com/terms", "kind": "terms", "status": 200, "checkedAt": "2026-11-07T14:01:27.593286996Z", "changedAt": "2026-11-03T07:43:22.480965239Z", "fingerprint": "b9ec671a2caf" } ], "updatedAt": "2026-11-07T22:55:51.650380142Z" } }, "answer": "Cloud Run GPUs scores 68.4 (B) on agent readiness against Together GPU Clusters's 58.1 (C), and leads in 5 of 7 scored categories. Together GPU Clusters leads on schema \u0026 documentation.", "b": { "slug": "together-gpu-clusters", "name": "Together GPU Clusters", "vendor": "Together AI", "vendorUrl": "https://www.together.ai", "kind": "http-api", "category": "gpu-compute", "summary": "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.", "url": "https://www.anchorterminal.com/tools/together-gpu-clusters", "markdownUrl": "https://www.anchorterminal.com/tools/together-gpu-clusters.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/together-gpu-clusters.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/together-gpu-clusters.json", "dns": "together-gpu-clusters.grade.anchorterminal.com", "repo": "https://github.com/togethercomputer/together-py", "license": "Proprietary service under Together's terms of service. The Python SDK and CLI are Apache-2.0 and the OpenAPI document is MIT", "transports": [], "remoteUrl": "https://api.together.ai/v1/compute/clusters", "packages": [ { "registry": "pypi", "name": "together" }, { "registry": "npm", "name": "together-ai" } ], "auth": "api-key", "authNotes": "Bearer `TOGETHER_API_KEY`, scoped to a project, revocable and optionally expiring, the same keys as Together's other APIs. Only a project admin's key returns the cluster's admin kubeconfig. Team members can reach the Kubernetes API through OIDC with their own identity provider, and SSH uses OIDC or uploaded keys.", "pricing": "usage", "pricingNotes": "On-demand from $3.99 per GPU-hour (H100) to $9.99 (B300); reserved 3 to 180 days from $3.19 to $7.99; preemptible (preview) from $1.99. Storage per GiB-month. No ingress or egress fees. Reservations are prepaid and non-refundable, and creation needs enough credit to cover them.", "priceSummary": "Pay per use", "where": "hosted", "x402": { "level": "no", "evidence": "No x402, MPP or L402 in the GPU clusters docs, billing pages or OpenAPI document (checked 2026-11-06). An exhausted balance returns 402 to scale requests with no payment requirements.", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-11-06" }, "docsUrl": "https://docs.together.ai/docs/gpu-clusters-overview", "llmsTxt": "https://docs.together.ai/llms.txt", "openapi": "https://github.com/togethercomputer/openapi", "capabilities": [ "compute.gpu", "compute.containers", "compute.batch" ], "tags": [ "hosted", "official", "usage", "prepaid", "openapi", "llms-txt", "kubernetes", "slurm", "python", "typescript", "cli", "h100", "b200" ], "lastRelease": "2026-10-22", "graded": true, "anchor": { "graded": true, "score": 58.1, "grade": "C", "agentReady": false, "rank": 768, "ranked": true, "rankOf": 1510, "categoryRank": 10, "methodology": "0.7", "run": "2026-10-01", "scores": { "ergonomics": 55, "maintenance": 75, "payments": 20, "reliability": 38, "schema": 92, "security": 68, "transparency": 65 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "medium", "date": "2026-11-06" }, "negative": 0, "verdict": "Multi-node H100 to B300 clusters on Kubernetes or Slurm, created and deleted headlessly through an OpenAPI-described API, a CLI with JSON output and two SDKs, at published GPU-hour prices. Clusters take 10 minutes to an hour to provision, can't be paused, and have no status component, SLA or idempotent create.", "bestFor": "Training, fine-tuning and batch jobs that need several 8-GPU nodes with InfiniBand, run by a pipeline that creates and deletes clusters.", "strengths": [ "Public OpenAPI 3.1, llms.txt and Markdown docs, with an automation guide for CI pipelines", "Published per GPU-hour prices for on-demand, reserved and preemptible capacity, with no egress fees", "Kubernetes or Slurm on the same hardware, with InfiniBand and a persistent shared volume", "Automatic health checks and node repair, approved by hand or run automatically", "Project-scoped keys, admin-only kubeconfig and OIDC access for team members" ], "weaknesses": [ "Provisioning takes 10 minutes to an hour, and a cluster can't be paused or scaled to zero", "No status page component, SLA or published rate limits for GPU clusters", "No idempotency key on creation, so a retried create can bill a second cluster", "No free trial; reservations are prepaid and non-refundable", "Volumes bill until deleted and outlive the clusters they were attached to" ], "agentNotes": [ "Use `tg clusters`, not `tg beta clusters`, which stopped working with CLI 3.0 on 22 October 2026", "Pass `--json` and `--non-interactive`, poll `tg clusters retrieve` until status is Ready, and delete the cluster in a cleanup step", "Check for an existing cluster by name before retrying a create, since creation has no idempotency key", "Read live availability from the regions API and prices from the pricing page; the regions response has no prices", "Use a project admin's key when the task needs the kubeconfig, since other roles get none" ], "metrics": { "kind": "remote", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "medium", "grade": "C", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 58.1 } ], "editorialScores": { "ergonomics": 55, "maintenance": 75, "payments": 20, "reliability": 38, "schema": 92, "security": 68, "transparency": 46 }, "provenanceScore": 83, "checked": { "categories": { "ergonomics": "2026-11-06", "maintenance": "2026-11-06", "payments": "2026-11-06", "reliability": "2026-11-06", "schema": "2026-11-06", "security": "2026-11-06", "transparency": "2026-11-06" }, "last": "2026-11-06", "oldest": "2026-11-06", "stale": false, "staleAfterDays": 90 } }, "connect": { "install": "uv tool install \"together[cli]\"", "http": "curl -H \"Authorization: Bearer $TOGETHER_API_KEY\" https://api.together.ai/v1/compute/clusters" }, "letme": { "capability": "https://letme.dev/compute.gpu", "tool": "https://letme.dev/together-gpu-clusters" }, "sameCompany": [ "together-inference", "together-fine-tuning", "together-code-sandbox" ], "area": "models", "unitPrices": [ { "item": "H100 on-demand", "unit": "gpu-hour", "usd": 3.99, "note": "reserved $3.69 to $3.19, preemptible $1.99" }, { "item": "H200 on-demand", "unit": "gpu-hour", "usd": 5.99, "note": "reserved $4.99 to $3.99, preemptible $2.99" }, { "item": "B200 on-demand", "unit": "gpu-hour", "usd": 8.19, "note": "reserved $7.99 to $6.79, preemptible $4.09" }, { "item": "B300 on-demand", "unit": "gpu-hour", "usd": 9.99, "note": "preemptible $4.99, reserved by sales" } ], "provenance": { "legalEntity": "Together Computer, Inc.", "domain": "together.ai", "domainRegistered": "", "endpointOnVendorDomain": true, "terms": "https://www.together.ai/terms-of-service", "privacy": "https://www.together.ai/privacy", "statusPage": "https://status.together.ai", "changelog": "https://docs.together.ai/docs/changelog", "securityTxt": "valid", "checked": "2026-11-06", "notes": [ "Legal entity, terms, privacy and security.txt as on the Together AI Inference listing, checked 1 November 2026, and not re-read for this listing.", "docs.together.ai/robots.txt disallows only /cdn-cgi/ and /_next/, with Content-Signal ai-train=yes, search=yes, ai-input=yes. www.together.ai/robots.txt allows all.", "status.together.ai has no component for GPU clusters on 6 November 2026." ], "score": 83 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/together-gpu-clusters.json", "live": { "slug": "together-gpu-clusters", "probe": { "target": "https://api.together.ai/v1/compute/clusters", "method": "get", "lastAt": "2026-11-07T22:56:27.360809678Z", "lastOk": true, "lastStatus": 401, "lastMs": 259, "lastNote": "asks for credentials", "authRequired": true, "uptime24h": 100, "uptime30d": 100, "p50ms24h": 492, "p95ms24h": 827, "samples24h": 244, "samples30d": 246, "days": [ { "date": "2026-11-06", "probes": 13, "ok": 13 }, { "date": "2026-11-07", "probes": 233, "ok": 233 } ] }, "versions": [ { "registry": "github", "name": "togethercomputer/together-py", "version": "v3.1.1", "released": "2026-11-04", "seenAt": "2026-11-07T00:12:15.100983119Z" }, { "registry": "npm", "name": "together-ai", "version": "1.0.0", "seenAt": "2026-11-07T00:12:15.067840103Z" }, { "registry": "pypi", "name": "together", "version": "3.1.1", "released": "2026-11-04", "seenAt": "2026-11-07T00:12:14.921893617Z" } ], "githubStars": 12, "npmWeekly": 99619, "pypiWeekly": 424792, "securityTxt": { "url": "https://together.ai/.well-known/security.txt", "state": "valid", "expires": "2028-09-30T00:00:00.000Z", "checkedAt": "2026-11-07T16:00:47.073630606Z" }, "llmsTxt": { "url": "https://docs.together.ai/llms.txt", "ok": true, "status": 200, "checkedAt": "2026-11-07T15:34:12.138384896Z" }, "updatedAt": "2026-11-07T22:56:27.360809678Z" } }, "edges": [ { "by": 32, "category": "reliability", "leads": "cloud-run-gpus", "name": "Reliability", "reasons": { "a": "Hosted reading. Google Cloud's status page with a JSON incident feed (20). In the 90 days from 8 August 2026 the feed names Cloud Run in three multi-product incidents, a service outage on 20 August (15:40 to 19:20 UTC, 27 products, global and us-west1), a network degradation in us-central1-b on 1 September and a europe-west2 disruption on 20 October. One major (10). Resource limits and GPU quota are published with numbers (3 GPUs a region by default), but no request rate for the Admin API was found on the quotas page (10). Google's client libraries retry with backoff and the Admin API takes etags, while no GPU-specific retry guidance was found (10). An SLA for GPU services, 99.95% with zonal redundancy and 99.5% without (10). GPU services generally available since 7 April 2025 (10). 20+10+10+10+10+10 = 70.", "b": "Read with the hosted lines. status.together.ai lists the website, the playground and serverless models with no component for GPU clusters, and a vendor page with no component for the product takes 15 in this run, as for Together's Code Sandbox and fine-tuning listings (15). No incident history for clusters can be read (5). No rate limits for the clusters API found (0). Together's docs give 429 and 503 backoff guidance for its APIs, per the inference listing's 1 November check, and the CLI has `--non-interactive` and `--json` for scripted retries, but creation has no idempotency key (8). No SLA for clusters found; the published 99% commitment covers provisioned inference throughput only (0). Generally available since 9 September 2025 and out of the CLI's beta paths since 22 October 2026, with preemptible nodes in preview (10). 15+5+0+8+0+10 = 38." }, "scores": { "a": 70, "b": 38 } }, { "by": 19, "category": "transparency", "leads": "cloud-run-gpus", "name": "Transparency \u0026 trust", "reasons": { "a": "Closed service under the Google Cloud terms, with Apache-2.0 client libraries (15). The Cloud Data Processing Addendum and the privacy notice cover customer data, per the Document AI listing's 1 November reading; no Cloud Run statement on GPU instance data after shutdown was found (22). The Cloud terms promise 12 months' notice before a backwards-incompatible API change, and driver versions are announced in the release notes (16). Regions per GPU type listed, with Google's subprocessor list and data-location terms (16). 15+22+16+16 = 69.", "b": "Closed service under Together's terms, with Apache-2.0 SDKs, which don't raise the line for a closed API (15). The privacy policy's data table covers GPU clusters, per the 4 November Code Sandbox check, and the billing page states when volumes are deleted after credit runs out, with no retention period for data on deleted clusters (15). The 3.0 breaking change came with a migration guide on the day, and reservations carry dated end-of-term emails, with no notice period for API changes found (8). Regions are listed through the API and US regions appear in examples; subprocessors and data locations for clusters were not established (8). 15+15+8+8 = 46." }, "scores": { "a": 84, "b": 65 } }, { "by": 11, "category": "schema", "leads": "together-gpu-clusters", "name": "Schema \u0026 documentation", "reasons": { "a": "API reading. A public Discovery document for the Cloud Run Admin API v2 (25). No llms.txt; Google's docs answer Markdown only at an unlinked `.md.txt` address, read as half, as for the Document AI listing (5). The GPU page says when to turn zonal redundancy on or off and the best-practices page when to bake models into images or load them from Cloud Storage (16). The GPU type is a free string in a node selector, with CPU and memory minimums stated in prose (9). gcloud, YAML, Terraform and console steps on each page, and errors follow Google's status model with no Cloud Run GPU error list (11). Versioned API (v1 and v2) and dated release notes (15). 25+5+16+9+11+15 = 81.", "b": "Public OpenAPI 3.1 for /v1/compute/clusters and shared volumes, MIT, embedded in every reference page (25). llms.txt and a Markdown twin of every page (10). The docs say when to choose Kubernetes or Slurm and each capacity type, and the management page carries instructions for agents on reporting live availability and prices (18). Enums for cluster type, billing type, GPU type and status, required fields marked (13). CLI, Python, TypeScript and console steps on each task, with 402 and 403 cases documented, though no error catalogue for the clusters API (11). Versioned SDKs and a dated changelog with cluster entries (15). 25+10+18+13+11+15 = 92." }, "scores": { "a": 81, "b": 92 } }, { "by": 11, "category": "ergonomics", "leads": "cloud-run-gpus", "name": "Agent ergonomics", "reasons": { "a": "Platform reading. One `gcloud run deploy` with `--gpu` and `--gpu-type` adds a GPU to a container, and responses carry the service object with `fieldMask` on updates (15). List calls page with `pageSize` and `pageToken` (15). Google's status codes and messages, with quota errors that name the limit (12). Revisions are immutable, traffic can be moved back, and etags guard updates, but no idempotency key on create (12). Client libraries in many languages and gcloud, though a GPU deploy needs instance-based billing, minimum CPU and memory and, past 3 GPUs, a quota request (12). 15+15+12+12+12 = 66.", "b": "Cluster objects are compact, and `--json` gives machine-readable output, but list calls take no field selection (15). No pagination or filters on the clusters list (8). 402 and 403 with readable messages such as \"Insufficient credits for cluster creation\", and troubleshooting for common failures (12). Status polling until Ready is documented and deletes run non-interactively, but there is no idempotency key on creation and a duplicate create bills a second cluster (8). Official Python and TypeScript SDKs and a CLI, though a create takes GPU type, region, driver and CUDA versions, cluster type and billing type (12). 15+8+12+8+12 = 55." }, "scores": { "a": 66, "b": 55 } }, { "by": 10, "category": "security", "leads": "cloud-run-gpus", "name": "Security \u0026 auth", "reasons": { "a": "OAuth 2.0 tokens from service accounts under IAM. Google's APIs accept `access_token` as a query parameter, the 10-point deduction other Google API listings take (20). IAM roles separate viewing, deploying and invoking, and services can require IAM authentication for every call (16). The platform runs your own container and returns its output (10). Cloud Audit Logs record Admin API calls, and request logs go to Cloud Logging (15). Google's vulnerability reward programme, security.txt and certifications (17). 20+16+10+15+17 = 78.", "b": "Project-scoped API keys, revocable with optional expiry, read as Together's other listings read the same key model (30). Admin and editor roles with no read-only role. Only admins get the admin kubeconfig, which has cluster-admin rights, and OIDC lets editors use kubectl under RBAC. Deletes run without confirmation when `--non-interactive` is passed (8). The API returns cluster state, not third-party content (10). Invoices itemise reserved, on-demand and storage charges, and no audit log for cluster operations was found (5). security.txt with a HackerOne route, per the 1 November check of the inference listing (15). 30+8+10+5+15 = 68." }, "scores": { "a": 78, "b": 68 } }, { "by": 5, "category": "maintenance", "leads": "cloud-run-gpus", "name": "Maintenance \u0026 community", "reasons": { "a": "Closed service with official SDKs. The newest GPU change is 30 October 2026, jobs with GPUs running up to 12 hours (30). Cloud Run release notes dated 4 November, 2 November, 30 October, 24 October and more in the 90 days (20). Release notes, Google Cloud support and public trackers for the client libraries, which were not read (10). Official client libraries in eight languages and gcloud (15). CI results not checked (5). 30+20+10+15+5 = 80.", "b": "Closed service. The newest cluster change is CLI and SDK 3.0 on 22 October 2026, 15 days before this check (30). Dated cluster entries in the 90 days from 8 August include 12, 18, 19, 24, 27 and 28 August, 10 September and 22 October (20). Dated changelog, a support portal and Discord, with responsiveness not tested (8). Python SDK 3.1 and TypeScript SDK 1.0 are current and cover clusters (15). SDK CI and dependencies weren't audited (2). 30+20+8+15+2 = 75." }, "scores": { "a": 80, "b": 75 } } ], "facts": [ { "a": "Platform", "b": "HTTP API", "name": "Kind" }, { "a": "Google Cloud", "b": "Together AI", "name": "Vendor" }, { "a": "https://run.googleapis.com", "b": "https://api.together.ai/v1/compute/clusters", "name": "Hosted endpoint" }, { "a": "", "b": "", "name": "Transports" }, { "a": "OAuth", "b": "API key", "name": "Auth" }, { "a": "Pay per use", "b": "Pay per use", "name": "Pricing" }, { "a": "no", "b": "no", "name": "x402" }, { "a": "Proprietary service under the Google Cloud Platform terms. Client libraries are Apache-2.0", "b": "Proprietary service under Together's terms of service. The Python SDK and CLI are Apache-2.0 and the OpenAPI document is MIT", "name": "Licence" }, { "a": "no", "b": "no", "name": "Read-only variant documented" }, { "a": "no", "b": "yes", "name": "llms.txt" }, { "a": "2026-10-30", "b": "2026-10-22", "name": "Last release" }, { "a": "2026-09-02", "b": "no date given", "name": "Terms last updated" }, { "a": "2026-09-28", "b": "no date given", "name": "Privacy policy last updated" }, { "a": "not found in the text", "b": "not found in the text", "name": "Customer content may train models" }, { "a": "not found in the text", "b": "not found in the text", "name": "Terms restrict automated access" }, { "a": "not found in the text", "b": "yes", "name": "Terms restrict benchmarking" }, { "a": "not found in the text", "b": "not found in the text", "name": "Terms or service can change without notice" }, { "a": "not found in the text", "b": "not found in the text", "name": "Arbitration or class-action waiver" } ], "faq": [ { "answer": "Cloud Run GPUs scores 68.4 (B) on agent readiness against Together GPU Clusters's 58.1 (C), and leads in 5 of 7 scored categories. Together GPU Clusters leads on schema \u0026 documentation.", "question": "Which is better for AI agents, Cloud Run GPUs or Together GPU Clusters?" }, { "answer": "Yes. Cloud Run GPUs has a hosted endpoint at https://run.googleapis.com and Together GPU Clusters at https://api.together.ai/v1/compute/clusters.", "question": "Can an agent call Cloud Run GPUs and Together GPU Clusters without installing anything?" } ], "goodFor": [ { "aheadOn": [ "Reliability, 70 against 38", "Agent ergonomics, 66 against 55", "Security \u0026 auth, 78 against 68", "Maintenance \u0026 community, 80 against 75", "Transparency \u0026 trust, 84 against 65" ], "also": null, "goodFor": "Serving small and mid-size open models, or batch jobs, from a container that should scale to zero.", "slug": "cloud-run-gpus", "watchFor": "One GPU per instance, L4 (24 GB) or RTX PRO 6000 (96 GB) only, so large models don't fit" }, { "aheadOn": [ "Schema \u0026 documentation, 92 against 81" ], "also": null, "goodFor": "Training, fine-tuning and batch jobs that need several 8-GPU nodes with InfiniBand, run by a pipeline that creates and deletes clusters.", "slug": "together-gpu-clusters", "watchFor": "Provisioning takes 10 minutes to an hour, and a cluster can't be paused or scaled to zero" } ], "history": null, "job": { "capability": "compute.gpu", "name": "GPU compute" }, "others": [ { "json": "https://www.anchorterminal.com/compare/baseten-vs-cloud-run-gpus.json", "title": "Baseten vs Cloud Run GPUs", "url": "https://www.anchorterminal.com/compare/baseten-vs-cloud-run-gpus" }, { "json": "https://www.anchorterminal.com/compare/baseten-vs-together-gpu-clusters.json", "title": "Baseten vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/baseten-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/beam-vs-cloud-run-gpus.json", "title": "Beam vs Cloud Run GPUs", "url": "https://www.anchorterminal.com/compare/beam-vs-cloud-run-gpus" }, { "json": "https://www.anchorterminal.com/compare/beam-vs-together-gpu-clusters.json", "title": "Beam vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/beam-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/cerebrium-vs-cloud-run-gpus.json", "title": "Cerebrium vs Cloud Run GPUs", "url": "https://www.anchorterminal.com/compare/cerebrium-vs-cloud-run-gpus" }, { "json": "https://www.anchorterminal.com/compare/cerebrium-vs-together-gpu-clusters.json", "title": "Cerebrium vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/cerebrium-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-coreweave.json", "title": "Cloud Run GPUs vs CoreWeave", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-coreweave" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-crusoe-cloud.json", "title": "Cloud Run GPUs vs Crusoe Cloud", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-crusoe-cloud" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-gpusmarket.json", "title": "Cloud Run GPUs vs GPUsMarket", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-gpusmarket" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-hugging-face-inference-endpoints.json", "title": "Cloud Run GPUs vs Hugging Face Inference Endpoints", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-hugging-face-inference-endpoints" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-hyperbolic.json", "title": "Cloud Run GPUs vs Hyperbolic", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-hyperbolic" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-lambda.json", "title": "Cloud Run GPUs vs Lambda Cloud", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-lambda" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-massed-compute.json", "title": "Cloud Run GPUs vs Massed Compute", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-massed-compute" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-nebius-ai-cloud.json", "title": "Cloud Run GPUs vs Nebius AI Cloud", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-nebius-ai-cloud" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-northflank.json", "title": "Cloud Run GPUs vs Northflank", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-northflank" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-thunder-compute.json", "title": "Cloud Run GPUs vs Thunder Compute", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-thunder-compute" }, { "json": "https://www.anchorterminal.com/compare/coreweave-vs-together-gpu-clusters.json", "title": "CoreWeave vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/coreweave-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/crusoe-cloud-vs-together-gpu-clusters.json", "title": "Crusoe Cloud vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/crusoe-cloud-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/gpusmarket-vs-together-gpu-clusters.json", "title": "GPUsMarket vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/gpusmarket-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/hugging-face-inference-endpoints-vs-together-gpu-clusters.json", "title": "Hugging Face Inference Endpoints vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/hugging-face-inference-endpoints-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/hyperbolic-vs-together-gpu-clusters.json", "title": "Hyperbolic vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/hyperbolic-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/koyeb-vs-together-gpu-clusters.json", "title": "Koyeb vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/koyeb-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/lambda-vs-together-gpu-clusters.json", "title": "Lambda Cloud vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/lambda-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/massed-compute-vs-together-gpu-clusters.json", "title": "Massed Compute vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/massed-compute-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/modal-vs-together-gpu-clusters.json", "title": "Modal vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/modal-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/nebius-ai-cloud-vs-together-gpu-clusters.json", "title": "Nebius AI Cloud vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/nebius-ai-cloud-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/northflank-vs-together-gpu-clusters.json", "title": "Northflank vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/northflank-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/replicate-deploy-vs-together-gpu-clusters.json", "title": "Replicate Deployments vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/replicate-deploy-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/runpod-vs-together-gpu-clusters.json", "title": "Runpod vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/runpod-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/thunder-compute-vs-together-gpu-clusters.json", "title": "Thunder Compute vs Together GPU Clusters", "url": "https://www.anchorterminal.com/compare/thunder-compute-vs-together-gpu-clusters" }, { "json": "https://www.anchorterminal.com/compare/together-gpu-clusters-vs-vast-ai.json", "title": "Together GPU Clusters vs Vast.ai", "url": "https://www.anchorterminal.com/compare/together-gpu-clusters-vs-vast-ai" }, { "json": "https://www.anchorterminal.com/compare/together-gpu-clusters-vs-verda.json", "title": "Together GPU Clusters vs Verda", "url": "https://www.anchorterminal.com/compare/together-gpu-clusters-vs-verda" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-koyeb.json", "title": "Cloud Run GPUs vs Koyeb", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-koyeb" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-modal.json", "title": "Cloud Run GPUs vs Modal", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-modal" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-replicate-deploy.json", "title": "Cloud Run GPUs vs Replicate Deployments", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-replicate-deploy" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-runpod.json", "title": "Cloud Run GPUs vs Runpod", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-runpod" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-vast-ai.json", "title": "Cloud Run GPUs vs Vast.ai", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-vast-ai" }, { "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-verda.json", "title": "Cloud Run GPUs vs Verda", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-verda" } ], "panel": null, "scores": [ { "by": 32, "cloud-run-gpus": 70, "edge": "cloud-run-gpus", "key": "reliability", "name": "Reliability", "together-gpu-clusters": 38, "weight": 16 }, { "key": "performance", "name": "Performance", "pending": true, "weight": 10 }, { "by": 11, "cloud-run-gpus": 81, "edge": "together-gpu-clusters", "key": "schema", "name": "Schema \u0026 documentation", "together-gpu-clusters": 92, "weight": 13 }, { "by": 11, "cloud-run-gpus": 66, "edge": "cloud-run-gpus", "key": "ergonomics", "name": "Agent ergonomics", "together-gpu-clusters": 55, "weight": 13 }, { "by": 10, "cloud-run-gpus": 78, "edge": "cloud-run-gpus", "key": "security", "name": "Security \u0026 auth", "together-gpu-clusters": 68, "weight": 14 }, { "by": 0, "cloud-run-gpus": 20, "edge": "", "key": "payments", "name": "Payments \u0026 pricing", "together-gpu-clusters": 20, "weight": 10 }, { "key": "tasks", "name": "Task success", "pending": true, "weight": 10 }, { "by": 5, "cloud-run-gpus": 80, "edge": "cloud-run-gpus", "key": "maintenance", "name": "Maintenance \u0026 community", "together-gpu-clusters": 75, "weight": 7 }, { "by": 19, "cloud-run-gpus": 84, "edge": "cloud-run-gpus", "key": "transparency", "name": "Transparency \u0026 trust", "together-gpu-clusters": 65, "weight": 7 } ], "summary": "Cloud Run GPUs scores 68.4 (B) on agent readiness against Together GPU Clusters's 58.1 (C), and leads in 5 of 7 scored categories. Together GPU Clusters leads on schema \u0026 documentation. Both do gpu compute.", "tasks": null, "tutorials": [], "verdicts": { "cloud-run-gpus": "Serverless GPUs on Cloud Run, an L4 or RTX PRO 6000 per instance, billed per second, scaling to zero and starting in about 5 seconds before the model loads, with a published SLA and IAM. One GPU per instance, 3 GPUs of default quota a region and instance-based billing limit it to small and mid-size models.", "together-gpu-clusters": "Multi-node H100 to B300 clusters on Kubernetes or Slurm, created and deleted headlessly through an OpenAPI-described API, a CLI with JSON output and two SDKs, at published GPU-hour prices. Clusters take 10 minutes to an hour to provision, can't be paused, and have no status component, SLA or idempotent create." } }, "kind": "anchor.page", "links": { "api": "https://www.anchorterminal.com/api/v1/index.json", "html": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-together-gpu-clusters", "json": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-together-gpu-clusters.json", "llms": "https://www.anchorterminal.com/llms.txt", "markdown": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-together-gpu-clusters.md", "mcp": "https://www.anchorterminal.com/mcp", "slim": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-together-gpu-clusters.min.md" }, "markdown": "Cloud Run GPUs scores 68.4 (B) on agent readiness against Together GPU Clusters's 58.1 (C), and leads in 5 of 7 scored categories. Together GPU Clusters leads on schema \u0026 documentation. Both do gpu compute.\n\n- Cloud Run GPUs: grade B, 68.4/100, rank #289 of 1510. Markdown https://www.anchorterminal.com/tools/cloud-run-gpus.md · JSON https://www.anchorterminal.com/api/v1/tools/cloud-run-gpus.json\n- Together GPU Clusters: grade C, 58.1/100, rank #768 of 1510. Markdown https://www.anchorterminal.com/tools/together-gpu-clusters.md · JSON https://www.anchorterminal.com/api/v1/tools/together-gpu-clusters.json\n- Best GPU and serverless compute for AI workloads: https://www.anchorterminal.com/best/gpu-compute/index.md\n- All 231 gpu compute comparisons: https://www.anchorterminal.com/compare/gpu-compute/index.md\n\n## Which one, for what\n\n### Cloud Run GPUs (B)\n\nGood for: Serving small and mid-size open models, or batch jobs, from a container that should scale to zero.\n\nAhead on:\n- Reliability, 70 against 38\n- Agent ergonomics, 66 against 55\n- Security \u0026 auth, 78 against 68\n- Maintenance \u0026 community, 80 against 75\n- Transparency \u0026 trust, 84 against 65\n\nWatch for: One GPU per instance, L4 (24 GB) or RTX PRO 6000 (96 GB) only, so large models don't fit\n\n### Together GPU Clusters (C)\n\nGood for: Training, fine-tuning and batch jobs that need several 8-GPU nodes with InfiniBand, run by a pipeline that creates and deletes clusters.\n\nAhead on:\n- Schema \u0026 documentation, 92 against 81\n\nWatch for: Provisioning takes 10 minutes to an hour, and a cluster can't be paused or scaled to zero\n\n\n## Why the edge\n\n### Reliability: Cloud Run GPUs +32\n\n- Cloud Run GPUs (70): Hosted reading. Google Cloud's status page with a JSON incident feed (20). In the 90 days from 8 August 2026 the feed names Cloud Run in three multi-product incidents, a service outage on 20 August (15:40 to 19:20 UTC, 27 products, global and us-west1), a network degradation in us-central1-b on 1 September and a europe-west2 disruption on 20 October. One major (10). Resource limits and GPU quota are published with numbers (3 GPUs a region by default), but no request rate for the Admin API was found on the quotas page (10). Google's client libraries retry with backoff and the Admin API takes etags, while no GPU-specific retry guidance was found (10). An SLA for GPU services, 99.95% with zonal redundancy and 99.5% without (10). GPU services generally available since 7 April 2025 (10). 20+10+10+10+10+10 = 70.\n- Together GPU Clusters (38): Read with the hosted lines. status.together.ai lists the website, the playground and serverless models with no component for GPU clusters, and a vendor page with no component for the product takes 15 in this run, as for Together's Code Sandbox and fine-tuning listings (15). No incident history for clusters can be read (5). No rate limits for the clusters API found (0). Together's docs give 429 and 503 backoff guidance for its APIs, per the inference listing's 1 November check, and the CLI has `--non-interactive` and `--json` for scripted retries, but creation has no idempotency key (8). No SLA for clusters found; the published 99% commitment covers provisioned inference throughput only (0). Generally available since 9 September 2025 and out of the CLI's beta paths since 22 October 2026, with preemptible nodes in preview (10). 15+5+0+8+0+10 = 38.\n\n### Transparency \u0026 trust: Cloud Run GPUs +19\n\n- Cloud Run GPUs (84): Closed service under the Google Cloud terms, with Apache-2.0 client libraries (15). The Cloud Data Processing Addendum and the privacy notice cover customer data, per the Document AI listing's 1 November reading; no Cloud Run statement on GPU instance data after shutdown was found (22). The Cloud terms promise 12 months' notice before a backwards-incompatible API change, and driver versions are announced in the release notes (16). Regions per GPU type listed, with Google's subprocessor list and data-location terms (16). 15+22+16+16 = 69.\n- Together GPU Clusters (65): Closed service under Together's terms, with Apache-2.0 SDKs, which don't raise the line for a closed API (15). The privacy policy's data table covers GPU clusters, per the 4 November Code Sandbox check, and the billing page states when volumes are deleted after credit runs out, with no retention period for data on deleted clusters (15). The 3.0 breaking change came with a migration guide on the day, and reservations carry dated end-of-term emails, with no notice period for API changes found (8). Regions are listed through the API and US regions appear in examples; subprocessors and data locations for clusters were not established (8). 15+15+8+8 = 46.\n\n### Schema \u0026 documentation: Together GPU Clusters +11\n\n- Cloud Run GPUs (81): API reading. A public Discovery document for the Cloud Run Admin API v2 (25). No llms.txt; Google's docs answer Markdown only at an unlinked `.md.txt` address, read as half, as for the Document AI listing (5). The GPU page says when to turn zonal redundancy on or off and the best-practices page when to bake models into images or load them from Cloud Storage (16). The GPU type is a free string in a node selector, with CPU and memory minimums stated in prose (9). gcloud, YAML, Terraform and console steps on each page, and errors follow Google's status model with no Cloud Run GPU error list (11). Versioned API (v1 and v2) and dated release notes (15). 25+5+16+9+11+15 = 81.\n- Together GPU Clusters (92): Public OpenAPI 3.1 for /v1/compute/clusters and shared volumes, MIT, embedded in every reference page (25). llms.txt and a Markdown twin of every page (10). The docs say when to choose Kubernetes or Slurm and each capacity type, and the management page carries instructions for agents on reporting live availability and prices (18). Enums for cluster type, billing type, GPU type and status, required fields marked (13). CLI, Python, TypeScript and console steps on each task, with 402 and 403 cases documented, though no error catalogue for the clusters API (11). Versioned SDKs and a dated changelog with cluster entries (15). 25+10+18+13+11+15 = 92.\n\n### Agent ergonomics: Cloud Run GPUs +11\n\n- Cloud Run GPUs (66): Platform reading. One `gcloud run deploy` with `--gpu` and `--gpu-type` adds a GPU to a container, and responses carry the service object with `fieldMask` on updates (15). List calls page with `pageSize` and `pageToken` (15). Google's status codes and messages, with quota errors that name the limit (12). Revisions are immutable, traffic can be moved back, and etags guard updates, but no idempotency key on create (12). Client libraries in many languages and gcloud, though a GPU deploy needs instance-based billing, minimum CPU and memory and, past 3 GPUs, a quota request (12). 15+15+12+12+12 = 66.\n- Together GPU Clusters (55): Cluster objects are compact, and `--json` gives machine-readable output, but list calls take no field selection (15). No pagination or filters on the clusters list (8). 402 and 403 with readable messages such as \"Insufficient credits for cluster creation\", and troubleshooting for common failures (12). Status polling until Ready is documented and deletes run non-interactively, but there is no idempotency key on creation and a duplicate create bills a second cluster (8). Official Python and TypeScript SDKs and a CLI, though a create takes GPU type, region, driver and CUDA versions, cluster type and billing type (12). 15+8+12+8+12 = 55.\n\n### Security \u0026 auth: Cloud Run GPUs +10\n\n- Cloud Run GPUs (78): OAuth 2.0 tokens from service accounts under IAM. Google's APIs accept `access_token` as a query parameter, the 10-point deduction other Google API listings take (20). IAM roles separate viewing, deploying and invoking, and services can require IAM authentication for every call (16). The platform runs your own container and returns its output (10). Cloud Audit Logs record Admin API calls, and request logs go to Cloud Logging (15). Google's vulnerability reward programme, security.txt and certifications (17). 20+16+10+15+17 = 78.\n- Together GPU Clusters (68): Project-scoped API keys, revocable with optional expiry, read as Together's other listings read the same key model (30). Admin and editor roles with no read-only role. Only admins get the admin kubeconfig, which has cluster-admin rights, and OIDC lets editors use kubectl under RBAC. Deletes run without confirmation when `--non-interactive` is passed (8). The API returns cluster state, not third-party content (10). Invoices itemise reserved, on-demand and storage charges, and no audit log for cluster operations was found (5). security.txt with a HackerOne route, per the 1 November check of the inference listing (15). 30+8+10+5+15 = 68.\n\n### Maintenance \u0026 community: Cloud Run GPUs +5\n\n- Cloud Run GPUs (80): Closed service with official SDKs. The newest GPU change is 30 October 2026, jobs with GPUs running up to 12 hours (30). Cloud Run release notes dated 4 November, 2 November, 30 October, 24 October and more in the 90 days (20). Release notes, Google Cloud support and public trackers for the client libraries, which were not read (10). Official client libraries in eight languages and gcloud (15). CI results not checked (5). 30+20+10+15+5 = 80.\n- Together GPU Clusters (75): Closed service. The newest cluster change is CLI and SDK 3.0 on 22 October 2026, 15 days before this check (30). Dated cluster entries in the 90 days from 8 August include 12, 18, 19, 24, 27 and 28 August, 10 September and 22 October (20). Dated changelog, a support portal and Discord, with responsiveness not tested (8). Python SDK 3.1 and TypeScript SDK 1.0 are current and cover clusters (15). SDK CI and dependencies weren't audited (2). 30+20+8+15+2 = 75.\n\n## Score by category\n\n| Category | Weight | Cloud Run GPUs | Together GPU Clusters | Edge |\n| --- | --- | --- | --- | --- |\n| Reliability | 16% (20 this run) | 70 | 38 | Cloud Run GPUs +32 |\n| Performance | 10%, pending | pending | pending | not scored in this run |\n| Schema \u0026 documentation | 13% (16.2 this run) | 81 | 92 | Together GPU Clusters +11 |\n| Agent ergonomics | 13% (16.2 this run) | 66 | 55 | Cloud Run GPUs +11 |\n| Security \u0026 auth | 14% (17.5 this run) | 78 | 68 | Cloud Run GPUs +10 |\n| Payments \u0026 pricing | 10% (12.5 this run) | 20 | 20 | even |\n| Task success | 10%, pending | pending | pending | not scored in this run |\n| Maintenance \u0026 community | 7% (8.8 this run) | 80 | 75 | Cloud Run GPUs +5 |\n| Transparency \u0026 trust | 7% (8.8 this run) | 84 | 65 | Cloud Run GPUs +19 |\n| Negative events | ≤15 | 0 | 0 | |\n| **Total** | | **68.4 · B** | **58.1 · C** | |\n\n## Facts side by side\n\n| Fact | Cloud Run GPUs | Together GPU Clusters |\n| --- | --- | --- |\n| Kind | Platform | HTTP API |\n| Vendor | Google Cloud | Together AI |\n| Hosted endpoint | `https://run.googleapis.com` | `https://api.together.ai/v1/compute/clusters` |\n| Transports | | |\n| Auth | OAuth | API key |\n| Pricing | Pay per use | Pay per use |\n| x402 | no | no |\n| Licence | Proprietary service under the Google Cloud Platform terms. Client libraries are Apache-2.0 | Proprietary service under Together's terms of service. The Python SDK and CLI are Apache-2.0 and the OpenAPI document is MIT |\n| Read-only variant documented | no | no |\n| llms.txt | no | yes |\n| Last release | 2026-10-30 | 2026-10-22 |\n| Terms last updated | 2026-09-02 | no date given |\n| Privacy policy last updated | 2026-09-28 | no date given |\n| Customer content may train models | not found in the text | not found in the text |\n| Terms restrict automated access | not found in the text | not found in the text |\n| Terms restrict benchmarking | not found in the text | yes |\n| Terms or service can change without notice | not found in the text | not found in the text |\n| Arbitration or class-action waiver | not found in the text | not found in the text |\n\n## Verdicts\n\n**Cloud Run GPUs.** Serverless GPUs on Cloud Run, an L4 or RTX PRO 6000 per instance, billed per second, scaling to zero and starting in about 5 seconds before the model loads, with a published SLA and IAM. One GPU per instance, 3 GPUs of default quota a region and instance-based billing limit it to small and mid-size models.\n\n**Together GPU Clusters.** Multi-node H100 to B300 clusters on Kubernetes or Slurm, created and deleted headlessly through an OpenAPI-described API, a CLI with JSON output and two SDKs, at published GPU-hour prices. Clusters take 10 minutes to an hour to provision, can't be paused, and have no status component, SLA or idempotent create.\n\n## Before you call either\n\n### Cloud Run GPUs\n\n1. Pass `--gpu 1 --gpu-type nvidia-l4` (or nvidia-rtx-pro-6000) with at least 4 vCPU and 16 GiB, and `--no-cpu-throttling` for instance-based billing\n2. Load model weights from Cloud Storage at startup in parallel; bake them into the image only below about 10 GB\n3. Turn zonal redundancy off for cheaper, more readily granted quota when failover isn't required\n4. Set minimum instances to 0 unless cold starts matter, since minimum instances bill at the full rate\n5. Send OAuth tokens in the `Authorization` header, never as `access_token` in the URL\n\n### Together GPU Clusters\n\n1. Use `tg clusters`, not `tg beta clusters`, which stopped working with CLI 3.0 on 22 October 2026\n2. Pass `--json` and `--non-interactive`, poll `tg clusters retrieve` until status is Ready, and delete the cluster in a cleanup step\n3. Check for an existing cluster by name before retrying a create, since creation has no idempotency key\n4. Read live availability from the regions API and prices from the pricing page; the regions response has no prices\n5. Use a project admin's key when the task needs the kubeconfig, since other roles get none\n\n## Questions\n\n### Which is better for AI agents, Cloud Run GPUs or Together GPU Clusters?\n\nCloud Run GPUs scores 68.4 (B) on agent readiness against Together GPU Clusters's 58.1 (C), and leads in 5 of 7 scored categories. Together GPU Clusters leads on schema \u0026 documentation.\n\n### Can an agent call Cloud Run GPUs and Together GPU Clusters without installing anything?\n\nYes. Cloud Run GPUs has a hosted endpoint at https://run.googleapis.com and Together GPU Clusters at https://api.together.ai/v1/compute/clusters.\n\n\n## For agents\n\n- This comparison as JSON: https://www.anchorterminal.com/compare/cloud-run-gpus-vs-together-gpu-clusters.json, and with the fewest tokens: https://www.anchorterminal.com/compare/cloud-run-gpus-vs-together-gpu-clusters.min.md\n- Over MCP at https://www.anchorterminal.com/mcp (no key): `compare_tools {\"a\": \"cloud-run-gpus\", \"b\": \"together-gpu-clusters\"}`. From a terminal: `anchor compare cloud-run-gpus together-gpu-clusters`\n- Each listing in full: https://www.anchorterminal.com/api/v1/tools/cloud-run-gpus.json and https://www.anchorterminal.com/api/v1/tools/together-gpu-clusters.json\n\n## Other comparisons with Cloud Run GPUs or Together GPU Clusters\n\n- [Baseten vs Cloud Run GPUs](https://www.anchorterminal.com/compare/baseten-vs-cloud-run-gpus.md)\n- [Baseten vs Together GPU Clusters](https://www.anchorterminal.com/compare/baseten-vs-together-gpu-clusters.md)\n- [Beam vs Cloud Run GPUs](https://www.anchorterminal.com/compare/beam-vs-cloud-run-gpus.md)\n- [Beam vs Together GPU Clusters](https://www.anchorterminal.com/compare/beam-vs-together-gpu-clusters.md)\n- [Cerebrium vs Cloud Run GPUs](https://www.anchorterminal.com/compare/cerebrium-vs-cloud-run-gpus.md)\n- [Cerebrium vs Together GPU Clusters](https://www.anchorterminal.com/compare/cerebrium-vs-together-gpu-clusters.md)\n- [Cloud Run GPUs vs CoreWeave](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-coreweave.md)\n- [Cloud Run GPUs vs Crusoe Cloud](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-crusoe-cloud.md)\n- [Cloud Run GPUs vs GPUsMarket](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-gpusmarket.md)\n- [Cloud Run GPUs vs Hugging Face Inference Endpoints](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-hugging-face-inference-endpoints.md)\n- [Cloud Run GPUs vs Hyperbolic](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-hyperbolic.md)\n- [Cloud Run GPUs vs Lambda Cloud](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-lambda.md)\n- [Cloud Run GPUs vs Massed Compute](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-massed-compute.md)\n- [Cloud Run GPUs vs Nebius AI Cloud](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-nebius-ai-cloud.md)\n- [Cloud Run GPUs vs Northflank](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-northflank.md)\n- [Cloud Run GPUs vs Thunder Compute](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-thunder-compute.md)\n- [CoreWeave vs Together GPU Clusters](https://www.anchorterminal.com/compare/coreweave-vs-together-gpu-clusters.md)\n- [Crusoe Cloud vs Together GPU Clusters](https://www.anchorterminal.com/compare/crusoe-cloud-vs-together-gpu-clusters.md)\n- [GPUsMarket vs Together GPU Clusters](https://www.anchorterminal.com/compare/gpusmarket-vs-together-gpu-clusters.md)\n- [Hugging Face Inference Endpoints vs Together GPU Clusters](https://www.anchorterminal.com/compare/hugging-face-inference-endpoints-vs-together-gpu-clusters.md)\n- [Hyperbolic vs Together GPU Clusters](https://www.anchorterminal.com/compare/hyperbolic-vs-together-gpu-clusters.md)\n- [Koyeb vs Together GPU Clusters](https://www.anchorterminal.com/compare/koyeb-vs-together-gpu-clusters.md)\n- [Lambda Cloud vs Together GPU Clusters](https://www.anchorterminal.com/compare/lambda-vs-together-gpu-clusters.md)\n- [Massed Compute vs Together GPU Clusters](https://www.anchorterminal.com/compare/massed-compute-vs-together-gpu-clusters.md)\n- [Modal vs Together GPU Clusters](https://www.anchorterminal.com/compare/modal-vs-together-gpu-clusters.md)\n- [Nebius AI Cloud vs Together GPU Clusters](https://www.anchorterminal.com/compare/nebius-ai-cloud-vs-together-gpu-clusters.md)\n- [Northflank vs Together GPU Clusters](https://www.anchorterminal.com/compare/northflank-vs-together-gpu-clusters.md)\n- [Replicate Deployments vs Together GPU Clusters](https://www.anchorterminal.com/compare/replicate-deploy-vs-together-gpu-clusters.md)\n- [Runpod vs Together GPU Clusters](https://www.anchorterminal.com/compare/runpod-vs-together-gpu-clusters.md)\n- [Thunder Compute vs Together GPU Clusters](https://www.anchorterminal.com/compare/thunder-compute-vs-together-gpu-clusters.md)\n- [Together GPU Clusters vs Vast.ai](https://www.anchorterminal.com/compare/together-gpu-clusters-vs-vast-ai.md)\n- [Together GPU Clusters vs Verda](https://www.anchorterminal.com/compare/together-gpu-clusters-vs-verda.md)\n- [Cloud Run GPUs vs Koyeb](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-koyeb.md)\n- [Cloud Run GPUs vs Modal](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-modal.md)\n- [Cloud Run GPUs vs Replicate Deployments](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-replicate-deploy.md)\n- [Cloud Run GPUs vs Runpod](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-runpod.md)\n- [Cloud Run GPUs vs Vast.ai](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-vast-ai.md)\n- [Cloud Run GPUs vs Verda](https://www.anchorterminal.com/compare/cloud-run-gpus-vs-verda.md)\n", "meta": { "attribution": "Anchor Terminal (https://www.anchorterminal.com)", "attributionRequired": true, "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)", "docs": "https://www.anchorterminal.com/docs/", "generatedAt": "2026-11-07", "license": "CC-BY-4.0", "method": "https://www.anchorterminal.com/benchmark/", "methodology": "0.7", "openapi": "https://www.anchorterminal.com/openapi.json", "preview": false, "run": "2026-10-01", "runLabel": "October 2026 research run", "terms": "https://www.anchorterminal.com/terms/#data" }, "page": { "breadcrumbs": [ { "name": "Home", "url": "https://www.anchorterminal.com/" }, { "name": "Compare", "url": "https://www.anchorterminal.com/compare/" }, { "name": "Cloud Run GPUs vs Together GPU Clusters", "url": "" } ], "description": "Cloud Run GPUs scores 68.4 (B) to Together GPU Clusters's 58.1 (C) for gpu compute. Prices, MCP, x402, uptime and agent notes side by side.", "facts": [ "Cloud Run GPUs B 68.4", "Together GPU Clusters C 58.1", "scores" ], "h1": "Cloud Run GPUs vs Together GPU Clusters", "image": "https://www.anchorterminal.com/assets/og/default.png", "path": "/compare/cloud-run-gpus-vs-together-gpu-clusters", "published": "2026-10-01", "section": "tools", "title": "Cloud Run GPUs vs Together GPU Clusters for AI agents (2026)", "toc": null, "updated": "2026-11-07", "url": "https://www.anchorterminal.com/compare/cloud-run-gpus-vs-together-gpu-clusters" }, "tokens": { "markdown": 5000, "slim": 2680 }, "version": 1 }