NOUMENOpen app
NOUMEN

stepfun · Open weights

Step 3.5 Flash

Added Feb 2, 2026Open weights

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Use in the app

StepFun's most capable open-source reasoning model with visible reasoning traces. Built on a sparse Mixture-of-Experts architecture with 196B total parameters and only 11B active per token, it achieves frontier-level performance in math, logic, and agentic coding while reaching up to 350 tokens/sec. Supports 256K context. NOTE: This model runs via StepFun, which may log and train on your prompts.

  • Reasoning
Input
Text
Output
Text
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.10
Output
$0.30

Reference prices as published by the model provider. API example

Specifications

Context window
262.1K
Max output
65.5K
Average output (7 days)
2.4K tokens
Longer than 94% of 468 models

Side by side

Step 3.5 Flash compared

Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Step 3.5 FlashstepfunThis modelStep 5 PreviewstepfunRemove
Providerstepfunstepfun
Discovery tierOpen weightsBalanced
AddedFeb 2, 2026Sep 20, 2026
Limits
Context window262.14K1MBest
Max output65.54K65.54K
Average output2.4K tokens1.5K tokens
Pricing, USD per 1M tokens
Input$0.10Best$1.00
Output$0.30Best$2.70
Cache read—$0.05
Performance
Request success (24h)100.0%100.0%
Benchmark indexes
Intelligence Index16.643.7Best
Capabilities
ReasoningYesYes
ToolsNoYes
VisionNoYes
Structured outputNoYes
Video inputNoYes
Parallel toolsNoYes
Reasoning effort—low, medium, high

Public evaluations

Benchmarks

Independent evaluation results, with how each score ranks among the models in the Noumen catalog.

  • Step 3.5 Flash
  • Step 5 Preview

Intelligence Index

  • Step 3.5 Flash16.6
  • Step 5 Preview43.7

Step 3.5 Flash: Better than 43% of 326 scored models

3.8Catalog models by score55.1

Agentic work

  • AutomationBench-AA

    Workflow automation, penalized for guardrail violations

    • Step 3.5 Flash: —
    • Step 5 Preview: 51.0%

  • AutomationBench-AA tasks completed

    Workflows finished end to end without a guardrail violation

    • Step 3.5 Flash: —
    • Step 5 Preview: 17.7%

  • AA-Briefcasev1.1

    Agentic knowledge work, rated head to head

    • Step 3.5 Flash: —
    • Step 5 Preview: 1,423 Elo

  • GDPval-AAv2.1

    Economically valuable occupational tasks, rated head to head

    • Step 3.5 Flash: —
    • Step 5 Preview: 1,587 Elo

Document reasoning

  • GDP.pdf

    Professional reasoning over PDF documents, all-pass rate

    • Step 3.5 Flash: —
    • Step 5 Preview: 14.8%

  • AA-LCRv1.1

    Reasoning across long documents

    • Step 3.5 Flash: 50.0%
    • Step 5 Preview: 88.3%Top

  • MLCR-AA

    Long-context reasoning over medical records

    • Step 3.5 Flash: —
    • Step 5 Preview: 18.3%

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • Step 3.5 Flash: 21.1%
    • Step 5 Preview: 46.5%Top

  • GPQA Diamond

    Graduate-level science questions

    • Step 3.5 Flash: 83.1%
    • Step 5 Preview: —

  • IFBench

    Following precise output instructions

    • Step 3.5 Flash: 64.6%
    • Step 5 Preview: —

Coding

  • Terminal-Benchv4.0

    Practical software engineering in a terminal

    • Step 3.5 Flash: —
    • Step 5 Preview: 33.3%

  • SciCode

    Scientific computing in Python

    • Step 3.5 Flash: —
    • Step 5 Preview: 58.9%

Legacy evaluations

  • Terminal-Bench Hard

    Earlier terminal task suite

    • Step 3.5 Flash: 27.3%
    • Step 5 Preview: —

  • τ²-Bench Telecom

    Tool use in a customer-support simulation

    • Step 3.5 Flash: 94.4%
    • Step 5 Preview: —

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • Step 3.5 Flash: 50.0%
    • Step 5 Preview: 88.3%Top

Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis

Live routing data

Performance and reliability

Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.

Prompt cache hits
0.0%
Request success (24h)
100.0%
Reliable

Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.

Hourly request success · last 72 hoursLowest hour 100.0%
  • Healthy (98% or more)
  • Degraded (90–98%)
  • Down (under 90%)
  • No requests

Developers

Use it through the API

Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.

Read the quickstart
curl https://www.noumen.xyz/api/v1/chat/completions \
  -H "Authorization: Bearer $NOUMEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "stepfun-ai/step-3.5-flash",
    "messages": [{ "role": "user", "content": "Explain this stack trace." }]
  }'
import os
from openai import OpenAI

client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
    model="stepfun-ai/step-3.5-flash",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

stepfun

Other stepfun models, starting with the same family.