RunPod Serverless GPU prices are up — B200 up about 25%, every tier higher

What changed — the new figures read on RunPod's own pricing page (runpod.io/pricing) on Nov 2, 2026, against the figures captured from the same page on Oct 26, 2026: B300 $10.65 → $11.99 (up about 13%), B200 $8.64 → $10.79 (up about 25%), H200 $5.93 → $6.99 (up about 18%), H100 $4.79 → $5.39 (up about 13%), RTX 5090 $1.58 → $1.79 (up about 13%), RTX 4090 $1.10 → $1.34 (up about 22%), and the grouped L4 / A5000 / 3090 rate $0.69 → $0.75 (up about 9%). Every Serverless tier rose; none fell. On sourcing, plainly: the after-figures are RunPod's own page as we read it today; the before-figures are PricingSaaS Pulse's dated Oct 26 capture of that same page, because our own archive does not hold the October Serverless buckets — the comparison is recorded at pricingsaas.com/companies/runpod/diffs/2026W45. There is no RunPod announcement and no stated effective date; the honest date for the change is the window between the two captures.

Also recorded between the same two captures: the Serverless line “Save 25% over other Serverless cloud providers on flex workers” was removed from the page, and the section now points to contract pricing instead of the long-term-commitment savings copy that sat there before.

What did not change: RunPod's Pods. Re-read on Nov 2, 2026, Pods Secure prices are exactly where they were — B300 $8.99, B200 $7.99, H200 $5.29, H100 SXM $3.99, A100 $1.79, L40S $1.09, RTX 5090 $1.19, RTX 4090 $0.89 and RTX PRO 6000 $2.49 an hour. Serverless is billed per second of worker time for inference endpoints, and its rates are not price-board rates, so the board does not move on this story — but anyone serving inference on RunPod Serverless is paying the new figures now.

What it costs: a Serverless worker that runs around the clock on a B200 now bills about $7,877 a month at $10.79 an hour, against about $6,307 at the old $8.64 — roughly $1,570 more per always-on GPU. An H100 worker running 24/7 goes from about $3,497 to about $3,935 a month. Serverless is normally bought for spiky traffic, where the per-second billing still beats an idle pod; the rises narrow that advantage at exactly the high end — B200 and H200 workers — where always-on endpoints are most common. This is RunPod's second rise in the recent record (Secure Cloud rose on Oct 14) and it extends the autumn pattern — Nebius on Oct 1, Verda eight times from Oct 23 to Nov 2, SaladCloud on Nov 1: rising prices are consistent with growing demand (our reading, labelled as interpretation; capacity and cost moves can also play a part).