Step 3.5 Flash 2603 is optimized for high-frequency agentic and coding workflows with improved token efficiency and faster reasoning. NOTE: This model runs via StepFun, which may log and train on your prompts.
Added Apr 14, 2026
Context Window
256.0K
Max Output
256.0K
Input Price (Auto)
$0.10/1M
Output Price (Auto)
$0.30/1M
Cache Read (Auto)
$0.050/1M
Capabilities
Benchmarks
Benchmarks
Performance metrics and benchmarks
Sourced from Artificial Analysis.
Reasoning
GPQA Diamond
Graduate-level scientific reasoning
82.6%
Better than 74% of models compared
HLE
Humanity's Last Exam
24.5%
Better than 80% of models compared
IFBench
Instruction-following benchmark
66.5%
Better than 79% of models compared
T²-Bench Telecom
Conversational AI agents in dual-control scenarios
87.4%
Better than 81% of models compared
AA-LCR
Long context reasoning evaluation
60.3%
Better than 63% of models compared
CritPt
Research-level physics reasoning
2.3%
Coding
SciCode
Python programming for scientific computing
38.5%
Better than 65% of models compared
Terminal-Bench Hard
Agentic coding and terminal use
32.6%
Better than 77% of models compared
Knowledge
AA-Omniscience Accuracy
Proportion of correctly answered questions
24.6%
AA-Omniscience Hallucination Rate
Rate of incorrect answers among non-correct responses
91.3%
Last updated Aug 7, 2026, 12:00 AM
Artificial AnalysisProviders
Auto routing is available for this model. Explicit provider selection is not available.
Loading provider options…
Related text models
Compare Step 3.5 Flash 2603 with similar models from the same provider or model family.
Step 3.5 Flash
stepfun-ai/step-3.5-flashStepFun'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.
Step 3.7 Flash Thinking
stepfun/step-3.7-flash:thinkingStep 3.7 Flash Thinking is StepFun's high-efficiency multimodal MoE model with visible reasoning enabled for deeper agentic coding, long-context reasoning, tool use, and native image/video understanding. ⚠️ Note: This model routes through StepFun, so privacy and logging guarantees may be limited.
Step R1 V Mini
step-r1-v-miniStep-R1-V-Mini, which supports image and text input, text output, has good instruction following and general capabilities, can perceive images with high precision and complete complex reasoning tasks.
Gemini 3.7 Flash
google/gemini-3.7-flashGoogle's frontier-performance Flash model for multimodal and agentic workloads, including coding, tool use, image understanding, PDF and document extraction, audio, and video. Google reports 65.3% on DeepSWE v1.1, up from 49.0% for Gemini 3.6 Flash, and 34% on GDP.pdf, up from 14%.
DeepSeek V4 Flash Latest
deepseek/deepseek-v4-flash-latestCompatibility alias that routes to the newest dated DeepSeek V4 Flash release. Currently uses Fireworks first for DeepSeek V4 Flash 0731, with automatic failover when needed. ⚠️ Privacy and logging guarantees are limited.
DeepSeek V4 Flash 0731 (Thinking)
deepseek/deepseek-v4-flash-0731:thinkingDeepSeek V4 Flash 0731 Thinking enables reasoning by default on the re-post-trained Mixture-of-Experts model with a 1M-token context window, built for coding, reasoning, and agent workflows.