comparison
GPU Flow vs Modal
Modal is a serverless platform: you write Python, decorate a function, and Modal runs it on a GPU, billing by the second. GPU Flow is three simpler things: a per-token inference API, a persistent sandbox with coding agents, and a dedicated B200 you reserve. Modal applies a multiplier when you pick a region; GPU Flow’s sandboxes and GPUs are in Spain, at one price. This page compares the three overlaps, with Modal’s prices as checked on 9 November 2026.
The short version
| Modal | GPU Flow | |
|---|---|---|
| Model | Serverless functions and sandboxes, billed per second of CPU, memory and GPU | Three products: token API, persistent sandbox per hour, dedicated GPU per hour or per month |
| Where it runs | If you pick a region, a multiplier on the base price: 1.15x for a broad region such as eu, 1.75x for a specific one such as eu-west, eu-north or eu-south | Sandboxes and dedicated GPUs in Spain, no regional surcharge |
| GPU | B200 $6.25/h, H200 $4.54/h, H100 $3.95/h, A100 80 GB $2.50/h (base price, before the region multiplier) | B200 only: MIG slices from €1.10 an hour and the full card at €6.50 an hour; reserved by the month, 10 % off |
| Sandboxes | $0.00003942 per core-second (one core is 2 vCPU) plus $0.00000667 per GiB-second; volumes at $0.09 per GiB-month, with 1 TiB a month free | Starter 4 vCPU / 8 GB at €0.15 an hour, Pro 8 vCPU / 32 GB at €0.30 an hour, Team 16 vCPU / 64 GB at €0.55 an hour; disk at €0.10 per GB-month, from 50 GB |
| Inference per token | Its pricing page has no per-token rate: you deploy your own model and pay GPU-seconds | 13 models, from €0.06 per million input tokens |
| Agents preinstalled | A container you set up yourself, with your own tools | Claude Code or OpenCode wired to the inference API over the internal network |
| Free credit | $30 a month on the free Starter plan; Team, $250 a month with $100 credit | €5 once, for Token Factory |
| Invoice and DPA | Prices in USD | EUR with VAT; data processing agreement (GDPR Article 28); ISO 27001 and ENS Medium |
Modal prices checked on 9 November 2026 on its pricing page and its region-selection guide, in USD and not converted. GPU Flow prices in EUR and live. Sources: modal.com/pricing, modal.com/docs/guide/region-selection
A B200 in Europe: what you actually pay
| Option | Per hour | Notes |
|---|---|---|
| Modal B200, base price | $6.25 | No region selected |
| Modal B200, eu region (1.15x) | ≈ $7.19 | Broad region: anywhere in the region |
| Modal B200, eu-west (1.75x) | ≈ $10.94 | Specific EU region |
| GPU Flow full B200, Spain | €6.50 | Dedicated, not serverless; €4,270.50 a month reserved (≈ €5.85 an hour) |
| GPU Flow 45 GB MIG slice | €1.95 | Hardware-isolated slice of a B200 |
If the GPU must stay in a specific EU country, the Modal row that counts is the specific-region one; at GPU Flow the card is in Spain and the monthly reservation takes 10 % off the hourly rate. For bursty, seconds-long jobs where the GPU sits idle most of the day, Modal’s per-second billing wins: you pay nothing between calls. For a model that has to be up all day, a reserved card wins.
A sandbox for an agent: an hour of work
A container with 4 vCPU and 8 GB running for one hour, from each platform’s unit prices.
| Option | One active hour | What is included |
|---|---|---|
| Modal Sandbox, 2 cores (4 vCPU) + 8 GiB | ≈ $0.48 | A container you set up yourself; volumes extra beyond the free tier; base price |
| GPU Flow Starter Sandbox, 4 vCPU + 8 GB | €0.15 | Claude Code or OpenCode preinstalled, persistent disk (€5.00 a month for 50 GB), auto-pause, SSH |
The comparison flips if your sandbox lives for seconds: Modal bills exactly those seconds, GPU Flow bills active hours with auto-pause after a minimum of 10 minutes. For a developer who works in the sandbox for hours, or an agent that runs for an afternoon, GPU Flow also comes with the agent already wired.
Inference: deploy it yourself or call it
On Modal you deploy an open-source model as a function and pay GPU-seconds while the container is up, including the time you keep it warm. On GPU Flow the model is already running; you pay per token. For a team that just needs Qwen3.8 27B or GLM 5.3 behind an API, the token price is simpler and, below a sustained high load, usually cheaper than keeping a GPU warm. For a custom or fine-tuned model that GPU Flow does not list, Modal (or a dedicated GPU Flow B200) is the way.
When Modal is the right choice
- 1Your workload is bursty, Python-native and lives in functions, not in long sessions.
- 2You need GPUs other than the B200, or many GPUs for a short time.
- 3You want to deploy your own model code and pay only for the seconds it runs.
When GPU Flow is the right choice
- 1You want sandboxes or GPUs in Spain, without a surcharge and without choosing a region.
- 2You want to call models, not deploy them: per-token pricing, OpenAI-compatible, 13 models ready.
- 3You want a sandbox where Claude Code is already installed and connected, for hours of work, with a disk that persists.
- 4You want a dedicated B200 slice or card, by the month, on an invoice in euros.
Frequently asked questions
Is GPU Flow a Modal alternative in Europe?
For sandboxes, dedicated GPUs and hosted inference, yes. For serverless Python functions you only pay for while they run, Modal does something GPU Flow does not.
Can I run my own model on GPU Flow?
Yes, on a dedicated B200 slice or card with SSH access. Not as a serverless function.
Does GPU Flow charge extra for EU regions?
No. Sandboxes and dedicated GPUs are in Spain, at one price, with no region multiplier.
Which is cheaper for an agent sandbox?
For sessions of an hour or more with 4 vCPU and 8 GB, GPU Flow: €0.15 an hour against ≈ $0.48 on Modal, each in its own currency. For sessions of seconds, Modal, which bills only those seconds.
Try it with €5 free
The €5 welcome credit is Token Factory balance: call any model in the catalogue with your OpenAI code. Sandboxes and GPUs are paid separately, by the hour.