Step 3.7 Flash

May 2026
Step 3.7 Flash
Compare in
on
—per task?
$0.20 / $1.15per 1M in / outcache $0.04
123 tok/sfirst token in 2.2 s?
256Kcontext

Rankingat high effort?

Strong at

S
Tool calling#4 of 242
B
Reasoning#119 of 334Science#111 of 312Long context#97 of 273Instructions#63 of 245Vision#48 of 124
C
Agents & work tasks#95 of 175Knowledge#166 of 318

Alternatives?

ModelScoreCostPer taskWhat changes
CheaperClaude Haiku 5.5max×2.2-54%$0.21Remove sampling parameters+5BetterGPT-6 Lunalow+10%-54%$0.005Function calling only with none · Chat Completions+3OpenGLM 5.3 Flashmax×2.1-46%$0.28Memory (4-bit): 121 GB → 193 GB+5CheaperMiniMax M3.1 Flashthinking×2.2-40%$0.22Complete model response must be appendedCheaperQwen3.8 FlashthinkingOpen×2-47%$0.37max_tokens range · QwenCloud+5

Run it yourself

estimate
Fits on✗ 8 GB (phone, laptop)✗ 16 GB✗ 24 GB GPU✗ 48 GB GPU✗ 80 GB GPU
LicenceApache-2.0free for commercial use

History

4 versions
171819Apr 26Release dateAA Intelligence Index 4.3.23.5 · Apr3.7
171819Apr 26Release dateAA Intelligence Index 4.3.23.5 · Apr3.7

Benchmarks

all categories
Chat & Knowledge
Humanity's Last Exam21.4#119
AA Omniscience Index-37.3#167
CritPt (physics)2.3#103
AA Intelligence Index 4.3.2≈19.5#123
AA Omniscience · Accuracy25.8#105
AA Omniscience · Hallucination Rate85#200
AA-LCR (long context) 1.173.7#97
GPQA Diamond80.9#112
Coding
SciCode43.9#78
Terminal-Bench Hard (version unstated)35.6#41
Terminal-Bench 2.139.3#86
Agents & Tool Use
GDPval-AA (work tasks) 2.117.8#95
τ³-Banking · AA12#83
ITBench SRE30.3#30
APEX-Agents14.8#21
EnterpriseOps-Gym-AA29.1#36
τ²-bench · AA98.5#4
Writing & Language
IFBench (instructions)67.3#63
Vision & Documents
MMMU-Pro (vision)75.3#48

FAQ

How much does Step 3.7 Flash cost?›

$0.20 per 1M input tokens and $1.15 per 1M output tokens; cached input $0.04 per 1M; cheapest via stepfun at $0.18 / $1.11.

Is Step 3.7 Flash free to use?›

There is no free API; the cheapest paid price is $0.18 per 1M input tokens. The weights are open, so you can run it yourself.

How good is Step 3.7 Flash, and what is it best at?›

#119 of 334 in Chat & Knowledge, #78 of 140 in Coding, #95 of 175 in Agents & Tool Use, #63 of 245 in Writing & Language, #48 of 124 in Vision & Documents. Strongest at: Tool calling.

Is Step 3.7 Flash better than Step 3.5 Flash?›

On AA Intelligence Index 4.3.2, Step 3.7 Flash scores 19.5 vs 17 — higher.

How can I use Step 3.7 Flash?›

Through the API of DeepInfra, Featherless, Novita, StepFun.

What is the context window of Step 3.7 Flash?›

256K tokens, with up to 230K output tokens.

When was Step 3.7 Flash released? Is there a newer version?›

Released 2026-05-29. It is the current version in its line.

Is Step 3.7 Flash open source?›

Yes, the weights are open (Apache-2.0 licence).

Can I run Step 3.7 Flash locally?›

The weights are open and can be self-hosted; hardware needs are not confirmed yet.

Can Step 3.7 Flash read images?›

It accepts text and images, video as input.

What is Step 3.7 Flash?›

Step 3.7 Flash is an open-weights language model from StepFun, released May 2026, the successor to Step 3.5 Flash.

What are the best alternatives to Step 3.7 Flash?›

Claude Haiku 5.5 (cheaper), GPT-6 Luna (better), GLM 5.3 Flash (open weights).