Modal (Grade A)为 Claude AI 提供的安全测试数据 AI 技能。A 等级。使用时请注意：用户需要远程执行隔离代码 --- 可以是可抛弃的容器，选用的 GPU 访问 (T4→H100)，或一个更安全的位置放置未受信任的/代码.

What are the limitations of using Modal for long‑running or high‑throughput workloads?

Modal sandboxes are designed for one‑off or short‑duration jobs; they add latency and cost compared to local execution and are not suited for continuously running services or high queries‑per‑second use cases. Ephemeral containers are billed for the full timeout period, so long‑running workloads become expensive quickly.

Asked by Quang Nguyen · Mar 7, 2026

Do I need a Modal account or to provision my own GPUs to run code?

No. Modal is a fully managed service; you can create and run sandboxes without a Modal account or owning any GPU hardware. You only pay for the compute you request, and the environment is provided as a disposable, isolated container.

Asked by Yaw Owusu · Feb 26, 2026

Can I use specific GPU models like H100 or A100, and how does that affect cost?

Yes, you can select GPU types such as T4, A100, or H100 when creating a sandbox. GPU choice directly impacts the price: an A100 with a 300 s timeout costs $0.2020, while a 24‑hour H100 sandbox costs $192 upfront. Shorter timeouts keep costs low, but any timeout >300 s incurs hourly billing for the entire requested duration.

Asked by Noelia Campos · Jan 14, 2026

How is Modal priced and what does 'pay per call in USDC' mean?

Modal charges a flat rate for sandbox creation when the timeout is 300 seconds or less (e.g., $0.0120 for CPU, $0.2020 for an A100 GPU). For timeouts above 300 seconds the cost switches to an hourly rate billed upfront for the full requested lifetime, and the amount is non‑refundable even if the job finishes early.

Asked by Lena Fischer · Jan 10, 2026