Technical transformation state
Transformation supported
Verified public open-weight release with authoritative checkpoint evidence.
StepFun / Step 3.7 Flash
Transformation options for Step 3.7 Flash
Validated per release from registry evidence, pricing components and runtime compatibility.
Architecture
MoE Transformer
Parameters: 198.00B · Context window: 262,144
Supported transformations
Quantization
Validated per release
Runtime packaging
Validated per release
Managed hosting
Validated per release
Distillation
Validated per release
GPU targeting
Validated per release
Evaluation and benchmarking
Validated per release
Download package
Validated per release
Quantization
BF16
Precision, memory, context and runtime fit are evaluated from release evidence before an order becomes available.
FP8
Precision, memory, context and runtime fit are evaluated from release evidence before an order becomes available.
NVFP4
Precision, memory, context and runtime fit are evaluated from release evidence before an order becomes available.
Published packages
Engineering ready
Step 3.7 Flash FP8 Production Package - Downloadable production build
StepFun · Step 3.7 Flash
Package profile: Downloadable production build
Transformation operations: FP8 Quantization, Evaluation and benchmarking
Target GPU: Hardware selected in the builder
Runtime packaging: transformers
Configured delivery: Download package
Delivery options: Download package / Host with FusionPointAI / Download + host
fp8transformers Download packageHost with FusionPointAIDownload + host
USD 106,700
One-time engineering
Engineering ready
Step 3.7 Flash FP8 Production Package - GPU-targeted hosted production
StepFun · Step 3.7 Flash
Package profile: GPU-targeted hosted production
Transformation operations: FP8 Quantization, Runtime packaging, Evaluation and benchmarking
Target GPU: NVIDIA H200 SXM 141GB
GPU count: 2
Runtime packaging: transformers
Configured delivery: Download + host
Delivery options: Download package / Host with FusionPointAI / Download + host
fp8GPU targeted2 × NVIDIA H200 SXM 141GBtransformersRuntime optimized Download packageHost with FusionPointAIDownload + host
USD 112,700
One-time engineering
Engineering ready
Step 3.7 Flash FP8 Production Package - GPU-targeted hosted production
StepFun · Step 3.7 Flash
Package profile: GPU-targeted hosted production
Transformation operations: FP8 Quantization, Runtime packaging, Evaluation and benchmarking
Target GPU: NVIDIA H100 SXM 80GB
GPU count: 4
Runtime packaging: transformers
Configured delivery: Download + host
Delivery options: Download package / Host with FusionPointAI / Download + host
fp8GPU targeted4 × NVIDIA H100 SXM 80GBtransformersRuntime optimized Download packageHost with FusionPointAIDownload + host
USD 112,700
One-time engineering
Engineering ready
Step 3.7 Flash FP8 Production Package - GPU-targeted hosted production
StepFun · Step 3.7 Flash
Package profile: GPU-targeted hosted production
Transformation operations: FP8 Quantization, Runtime packaging, Evaluation and benchmarking
Target GPU: NVIDIA H100 PCIe 80GB
GPU count: 4
Runtime packaging: transformers
Configured delivery: Download + host
Delivery options: Download package / Host with FusionPointAI / Download + host
fp8GPU targeted4 × NVIDIA H100 PCIe 80GBtransformersRuntime optimized Download packageHost with FusionPointAIDownload + host
USD 112,700
One-time engineering
Engineering ready
Step 3.7 Flash FP8 Production Package - GPU-targeted hosted production
StepFun · Step 3.7 Flash
Package profile: GPU-targeted hosted production
Transformation operations: FP8 Quantization, Runtime packaging, Evaluation and benchmarking
Target GPU: NVIDIA A100 SXM 80GB
GPU count: 4
Runtime packaging: transformers
Configured delivery: Download + host
Delivery options: Download package / Host with FusionPointAI / Download + host
fp8GPU targeted4 × NVIDIA A100 SXM 80GBtransformersRuntime optimized Download packageHost with FusionPointAIDownload + host
USD 112,700
One-time engineering
Engineering ready
Step 3.7 Flash FP8 Production Package - GPU-targeted hosted production
StepFun · Step 3.7 Flash
Package profile: GPU-targeted hosted production
Transformation operations: FP8 Quantization, Runtime packaging, Evaluation and benchmarking
Target GPU: NVIDIA L40S 48GB
GPU count: 6
Runtime packaging: transformers
Configured delivery: Download + host
Delivery options: Download package / Host with FusionPointAI / Download + host
fp8GPU targeted6 × NVIDIA L40S 48GBtransformersRuntime optimized Download packageHost with FusionPointAIDownload + host
USD 112,700
One-time engineering
Design validated
Production distillation package for Step 3.7 Flash - Quality-retention distillation
StepFun · Step 3.7 Flash
Package profile: Quality-retention distillation
Transformation operations: Distillation, Evaluation and benchmarking
Target GPU: Hardware selected in the builder
Runtime packaging: transformers
Configured delivery: Download package
Delivery options: Download package / Host with FusionPointAI / Download + host
DistillationQuality retentiontransformers Download packageHost with FusionPointAIDownload + host
USD 240,363
One-time engineering
Design validated
Production distillation package for Step 3.7 Flash - Throughput-optimized distillation
StepFun · Step 3.7 Flash
Package profile: Throughput-optimized distillation
Transformation operations: Distillation, Evaluation and benchmarking
Target GPU: Hardware selected in the builder
Runtime packaging: transformers
Configured delivery: Download package
Delivery options: Download package / Host with FusionPointAI / Download + host
DistillationThroughput optimizedtransformers Download packageHost with FusionPointAIDownload + host
USD 240,363
One-time engineering