StepFun
StepFun
Multimodal model · China ·May 2026

Step 3.7 Flash

Multimodal Open Weights
198B
Parameters
256K
Context
3
Benchmarks

About Step 3.7 Flash

Step 3.7 Flash is an AI model developed by StepFun, in the multimodal category, released in 2026, made available as an open-weights model with 198B params. An open-weights (Apache 2.0) 198B sparse Mixture-of-Experts vision-language model pairing a 196B language backbone with .

On this page you'll find Step 3.7 Flash's full specifications, including a context window of 256K tokens, 3 benchmark results. Review provider pricing and benchmark scores below, or compare Step 3.7 Flash head-to-head with other multimodal models.

→ More models by StepFun → All multimodal models → AI benchmarks leaderboard

Run Step 3.7 Flash locally — what to buy

Share this build ↗

Sized to what Step 3.7 Flash actually needs (~256k context at Q4), not to the biggest GPU: Good = cheapest that runs it, Better = best value, Best = most headroom. Speed is a hardware estimate; anything past ~120 tok/s is shown as instant.

Good
AMD Ryzen AI Max+ 395
Consumer GPU · ~$1,999
~1.8 tok/s
Q4 · tight on context · up to 5k ctx
Runs with: ExLlamaV2, KoboldCpp, LM Studio
See full estimate →
Better
Apple M3 Max 128GB
Apple · ~$4,700
~3 tok/s
Q4 · tight on context · up to 5k ctx
Runs with: ExLlamaV2, KoboldCpp, LM Studio
See full estimate →
Best
Apple M1 Ultra 128GB
Apple · ~$4,800
~5.9 tok/s
Q4 · tight on context · up to 5k ctx
Runs with: ExLlamaV2, KoboldCpp, LM Studio
See full estimate →

Software support

·ExLlamaV2 ·KoboldCpp ·LM Studio ·MLX ·Ollama ·SGLang ·TensorRT-LLM ·llama.cpp ·vLLM ?KTransformers ?text-generation-webui

✓ measured · · compatible · — not supported. Informational only — speed is hardware-based.

Consumer GPU
1.8–2.1 tok/s
2 options →
Apple
3–5.9 tok/s
4 options →
Datacenter GPU
24–62.8 tok/s
5 options →

Community

Would you run Step 3.7 Flash again?
No votes yet — be the first.
Did it run? — community reports by hardware

No reports yet. Own this model on some hardware? Be the first to confirm it runs.

Discussion

No comments yet. Start the discussion.

Step 3.7 Flash benchmark scores▶
Compare on the benchmark table →
Benchmark Category Score Variant
BrowseComp agentic 75.8%
Terminal-Bench 2.1 agentic 59.55%
SWE-bench Pro code 56.26%

More models like Step 3.7 Flash

AI
NeuroChess
· 72.3K
 StepFun
StepAudio 2.5 Realtime
StepFun
 StepFun
Step Image Edit 2
StepFun · 3.5B
 StepFun
Step-Audio-R1.5
StepFun
 StepFun
Step-Video-T2V
StepFun · 30B
 StepFun
Step-1
StepFun · 130B

Frequently asked questions

What is Step 3.7 Flash? ▶

An open-weights (Apache 2.0) 198B sparse Mixture-of-Experts vision-language model pairing a 196B language backbone with It is tracked on GenAIList with its specifications, benchmark scores and provider pricing.

Who created Step 3.7 Flash? ▶

Step 3.7 Flash was developed by StepFun and released in 2026.

Is Step 3.7 Flash open source or proprietary? ▶

Step 3.7 Flash ships as an open-weights model: you can download and self-host the weights, though the licence may place some restrictions on use.

What is Step 3.7 Flash's context window? ▶

Step 3.7 Flash supports a context window of 256K tokens, and has 198B params.

How much does Step 3.7 Flash cost? ▶

Pricing for Step 3.7 Flash depends on the provider. See the providers table on this page for the latest API rates.

How does Step 3.7 Flash perform on benchmarks? ▶

Step 3.7 Flash is benchmarked across 3 evaluations on GenAIList, including BrowseComp (75.8%). See the full benchmark table below and compare it with other models.

Step-3.7 Flash FREE Unlimited API + Hermes Agent
Step 3.7 Flash First Look & LOCAL Test – A VERY Creative Model!
Step 3.7 Flash: The Open-Source AI for AI Agents

Know a great video about Step 3.7 Flash?