{"id":"production-inference-fleet","slug":"production-inference-fleet","name":"Self-hosted production inference fleet","shortName":"Self-hosted production inference fleet","problemStatement":"Always-on model serving on rented GPUs with control over stacks.","whenNotTopPick":"Prefer tokens not GPUs — Fireworks/Together. Huge enterprise K8s — CoreWeave/AWS.","categoryIds":["cat-gpu-cloud"],"depotPickProviderId":"runpod","depotPickRationale":"reliability + dx + pricePerf balance.","runnersUpIds":["coreweave","modal","baseten","aws-gpu","friendli"],"weights":{"capabilityFit":0.25,"pricePerformance":0.18,"latencyReliability":0.15,"developerDx":0.15,"privacyPosture":0.1,"lockInRisk":0.07,"maturity":0.1},"methodologyNotes":"reliability + dx + pricePerf balance.","tags":["production"],"depth":{"whenNotToUse":{"text":"Prefer tokens not GPUs — Fireworks/Together. Huge enterprise K8s — CoreWeave/AWS.","source":"editorial `whenNotTopPick` (data/jtbds.json)"},"whenNotToUseReason":null,"runnerUp":{"providerId":"coreweave","name":"CoreWeave","overallForJob":7.71,"pickOverallForJob":8.32,"beatsPickOn":["capabilityFit","latencyReliability","privacyPosture","maturity"],"source":"first resolvable entry of the job's editorially ranked `runnersUpIds`; scores are stored job-weighted scores"},"runnerUpReason":null,"lockIn":{"pickLockInRisk":3.5,"runnerUpLockInRisk":6.5,"scale":"0–10 stored curated estimate; higher = more lock-in (worse)","pickExitSignals":[],"runnerUpExitSignals":[],"pickNotes":[],"selfHostFallback":[{"id":"vllm","name":"vLLM","license":"Apache-2.0","sourceUrl":"https://github.com/vllm-project/vllm/blob/main/LICENSE"}],"migration":"The pick has lower stored lock-in risk than runner-up CoreWeave (3.5 vs 6.5). Self-host fallback listed for this job: vLLM.","migrationReason":null,"derivedFrom":"stored lockInRisk scores, provider tags (oss, self-host, openai-compat, open-models), free rows that list this job, and verbatim provider notes. No migration guide was invented."}},"freshness":{"pick":{"score":{"asOf":"2026-10-19","ageDays":8,"stale":false,"staleAfterDays":30,"reason":null},"lastReviewed":{"asOf":"2026-10-19","ageDays":8,"stale":false,"staleAfterDays":30,"reason":null},"sourceCheck":{"checkedAt":"2026-10-27","sourceUrls":2,"sourcesReachable":2,"pricingRows":12,"pricingRowsReconfirmed":0,"note":"A source re-check confirms the page is reachable and, for pricing rows, that the stored amount and SKU name appear on it. It does not re-derive or change the score.","reason":null}}},"depotPick":{"id":"runpod","slug":"runpod","name":"RunPod","tagline":"Depot default for self-serve GPUs. Secure Cloud for anything customer-facing; Community/spot for batch.","overall":8.32},"runnersUp":[{"id":"coreweave","slug":"coreweave","name":"CoreWeave","tagline":"Enterprise training/inference fleet pick. Skip if you just need a single H100 for an afternoon — use RunPod.","overall":7.71},{"id":"modal","slug":"modal","name":"Modal","tagline":"Depot darling for Python agent backends and bursty GPU jobs. If you think in functions not VMs, start here.","overall":8.44},{"id":"baseten","slug":"baseten","name":"Baseten","tagline":"Strong custom-model serving alternative to Replicate/Modal when you bring your own model.","overall":7.99},{"id":"aws-gpu","slug":"aws-gpu","name":"AWS GPU (EC2/P5/SageMaker)","tagline":"Use when AWS is destiny. For greenfield agent GPUs, neoclouds win on price and speed-to-pod.","overall":7.72},{"id":"friendli","slug":"friendli","name":"FriendliAI","tagline":"Open-model production peer when latency/tool-calling matter. Fireworks remains depot open-model default…","overall":8.03}]}