Step 3.5 Flash

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

Added Feb 2, 2026

Model weights

Pricing

Auto routing · per 1M tokens

Input: $0.10
Output: $0.30

Compare provider prices

Specifications

Context window: 256K
Max output: 256K
Parameters: 196B / 11B; Total / active
Avg output (7d): 1K tokens; Longer than 71% of models

Benchmarks

Artificial Analysis

LMArena

Sourced from Artificial Analysis.

Intelligence Index

16.6

Better than 59% of models compared

Agentic work

T²-Bench Telecom (legacy)

Legacy fallback · Conversational AI agents in dual-control scenarios

94.4%

Better than 92% of models compared

Document reasoning

AA-LCR v1.1

Long context reasoning with updated grading

50.0%

Better than 44% of models compared

Reasoning

HLE

Humanity's Last Exam

21.1%

Better than 69% of models compared

IFBench

Instruction-following benchmark

64.6%

Better than 76% of models compared

CritPt

Research-level physics reasoning

2.5%

Coding

Terminal-Bench Hard (legacy)

Legacy fallback · Agentic coding and terminal use

27.3%

Better than 69% of models compared

Knowledge

AA-Omniscience Accuracy

Proportion of correctly answered questions

23.6%

AA-Omniscience Hallucination Rate

Rate of incorrect answers among non-correct responses

85.7%

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

83.1%

Better than 73% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation