Infrastructure execution for AI workloads
Build what’s next.
We handle
the compute.
Your command in. Your results out.Illustrated workflow
Your next workload starts here.
Fine-tuning · your training code
Training
Run pre-training, post-training, and fine-tuning, including LoRA, with your own code and data.
A place for your agent’s toolsresearch-agent EXAMPLE SESSION
sandbox.exec(...) command
iter_output() stdout
write(...) stdin
Example evaluation tasks
✓
✓
✓
×
✓
✓
✓
✓
✓
✓
✓
×
✓
✓
✓
✓
✓
✓
×
✓
✓
·
·
·
TASKS / OUTCOMES / REPORTS
Long-running agent workresearch-agentExecute toolsObserve output
Working within your time and spending limits
Agents
Sandbox previewRun agents in sandboxes, execute your evals, and support longer tasks with configurable limits.
Explore. Train. Evaluate.
Reinforcement learning
Explore task environments, run training templates, and compare evaluation results.
Inference
Coming soonA planned home for running models on new inputs, from batch predictions to interactive responses.
NODUS / COMPUTE COSTS
GPU and CPU compute costs.
16 GPU models. 6 CPU sizes. Modal comparisons where listed.
$2.600/ hourH100 SXM · 80 GB VRAM
Nodus supply$2.600/h
Modal$3.949/h
34%lower supply cost
$0.0357/ hour2 vCPU · 4 GiB RAM · 80 GiB disk
Nodus supply$0.0357/h
Modal standard$0.0791/h
55%lower supply cost
Built to keep your work moving.
Backed byY Combinator