Nodus
面向长时间GPU模型训练任务的云平台，提供小时供应价格计费、自动检查点保存与机器回收恢复功能
Timeline (last 90 days)
Nodus 平台在机器被回收时会自动在新机器上恢复检查点，而不是从头开始。
Nodus 的 Function 也有 24 小时上限，但长时间训练通常以 Job 方式运行，Job 的墙钟超时可调整或移除，暂停时间不计入。
Nodus's multi-node training is in beta, with gangs of up to 8 nodes, requiring a first credit purchase; by default it uses a private network of the same provider within the same region, and although cross-provider is allowed, during beta traffic over public addresses is unencrypted.
In Nodus's pilot on October 11, 2026, using a Qwen2.5-0.5B model running the same recipe on a single NVIDIA L4, supervised fine-tuning cost 67% less than Modal and continued pretraining cost 42.5% less; after counting one failed attempt, the fine-tuning savings dropped to 24.7%.
As of October 12, 2026, the Nodus pricing page shows hourly supply prices: H100 $2.60, A100 80GB $1.09, L4 $0.43.
Nodus 运行一个 MCP 服务器，Claude Code、Codex 和 Cursor 可以直接调用，每次写入以干运行形式返回成本估算，确认后才运行。
At Modal's list prices, running an 8×H100 machine continuously for one week (168 hours) costs $1,813 more than Nodus.