StepFun · Efficient heavy model

Host Step 3.7 Flash

A throughput-oriented multimodal agent model with only about 11B parameters active per generated token.

Customize workload
Parameters
201B
Context
262,144 tokens
Baseline
120GB VRAM
Quantization
NVFP4 / Q4 inference

Current hosting recommendations

Default interactive workload, 8K context, one concurrent request, always on.

14 qualifying tracked rows
Lowest cost 92.6/100

H200 on Lambda

$2.29/hr$1,672/mo
Memory
141GB per GPU
Pricing
on-demand
Evidence
75.0/100
Operations fit
85/100

141GB per GPU clears the 120GB planning floor.

Inventory breadth can be narrower, and unavailable GPUs may not produce current price rows.

Check Lambda View price detail Adjust plan
Balanced fit 86.1/100

MI300X on RunPod

$2.39/hr$1,745/mo
Memory
192GB per GPU
Pricing
on-demand
Evidence
75.0/100
Operations fit
85/100

192GB per GPU clears the 120GB planning floor.

Template availability and community or spot pricing can move quickly, so freshness matters.

Check RunPod View price detail Adjust plan
Production 82.1/100

MI300X on RunPod

$2.39/hr$1,745/mo
Memory
192GB per GPU
Pricing
on-demand
Evidence
75.0/100
Operations fit
85/100

192GB per GPU clears the 120GB planning floor.

Template availability and community or spot pricing can move quickly, so freshness matters.

Check RunPod View price detail Adjust plan

Planning floor

A planning floor derived from catalog VRAM, quantization, context, concurrency, and traffic headroom; benchmark the final runtime before purchase.

Total VRAM
120GB
Per GPU
120GB
GPU count
1

Continue the decision

Change traffic and uptimeRecalculate cost and headroom for your workload. Step 3.7 Flash on RunPodInspect provider-specific qualifying rows. Compare other modelsReview quality, memory, and hosting envelopes.
Agent-assisted setup

Hand this runbook to Claude Code or Codex

Open a terminal in the repository where you want the deployment files, start claude or codex, then paste this prompt. It asks the agent to verify sources and stop before it creates billable infrastructure.

Claude Code quickstart ↗ Codex docs ↗
Deployment prompt 
Use the infrastructure or model context on this page to create a reproducible open-model deployment. Use this guide as the starting context: https://www.getflops.ai/llms/step-3.7-flash. Read the linked model card and provider documentation before choosing hardware or runtime settings. Open every linked primary source and flag any mismatch instead of guessing. Create a deployment folder containing README.md, .env.example with no secrets, a pinned start script or infrastructure manifest, and smoke-test.sh. Make the endpoint OpenAI-compatible where the runtime supports it. Run local/static validation, estimate the billable resources, and stop before provisioning paid infrastructure until I approve.
Guardrails included No secrets in files · verify primary docs · approval before spend