StepFun 3.5 Flash
StepFun 3.5 Flash is an open-source sparse MoE reasoning model from StepFun with 196B total parameters and about 11B active per token. It supports a context window of 262.1K tokens and a max output of 262.1K tokens per request.
import { streamText } from 'ai'
const result = streamText({ model: 'stepfun/step-3.5-flash', prompt: 'Why is the sky blue?'})Playground
Try out StepFun 3.5 Flash by StepFun. Usage is billed to your team at API rates. Free users (those who haven't made a payment) get $5 of credits every 30 days.
StepFun 3.5 Flash
Ask StepFun 3.5 Flash anything to try it out.
Providers
Route requests across multiple providers. Copy a provider slug to set your preference. Visit the docs for more info. Using a provider means you agree to their terms, listed under Legal.
| Provider |
|---|
 DeepInfra | 262K | 0.6s | 106tps | $0.09/M | $0.30/M | Read:$0.02/M Write:— | — | — | +1 | 01/29/2026 |
P50 throughput on live AI Gateway traffic, in tokens per second (TPS). Visit the docs for more info.
P50 time to first token (TTFT) on live AI Gateway traffic, in milliseconds. View the docs for more info.
Direct request success rate on AI Gateway and per-provider. Visit the docs for more info.
More models by StepFun
| Model |
|---|
| 256K | 9.9s | 171tps | $0.20/M | $1.15/M | Read:$0.04/M Write:— | — | — | +2 | | 05/28/2026 |
About StepFun 3.5 Flash
StepFun 3.5 Flash is StepFun's open-source reasoning model, released under the Apache 2.0 license with weights published on GitHub and Hugging Face. The architecture is a sparse mixture of experts with 196B total parameters and about 11B active per token. Each token routes through eight of 288 experts plus one shared expert, so inference cost tracks the active subset rather than the full parameter count.
Two design choices keep long-context work affordable. A 3:1 ratio of sliding-window to full attention layers supports the 262.1K tokens context window without quadratic cost across every layer. Multi-token prediction lets StepFun 3.5 Flash draft several tokens per forward pass, which accelerates generation.
StepFun 3.5 Flash posts frontier-level reasoning scores for its size: 97.3 on AIME 2025, 74.4% on SWE-bench Verified, 86.4% on LiveCodeBench-v6, and 88.2% on Tau2-Bench. Deep reasoning, tool calling, and agentic control loops are first-class capabilities rather than add-ons.
Through AI Gateway, you call StepFun 3.5 Flash with a single API key and get provider routing, automatic failover, and built-in observability. Integrate via the AI SDK, the Chat Completions API, the Responses API, the Messages API, or other supported API formats. No StepFun account is required.
What To Consider When Choosing a Provider
- Configuration: StepFun 3.5 Flash is a reasoning model, so responses include thinking tokens before the final answer. Budget output tokens accordingly, and tune reasoning depth where your harness allows it.
- Configuration: StepFun 3.5 Flash is text-only. Route requests that carry images or video to a multimodal model instead. Benchmark figures above come from StepFun's published evaluations, so validate on your own workload before committing production traffic. For current throughput and latency numbers, see live metrics on this page.
- Zero Data Retention: AI Gateway supports Zero Data Retention for this model via direct gateway requests (BYOK is not included). To configure this, check the documentation.
- Authentication: AI Gateway authenticates requests using an API key or OIDC token. You do not need to manage provider credentials directly.
When to Use StepFun 3.5 Flash
Best For
- Agentic Tool Loops: Pipelines that chain deep reasoning with multi-step tool calls
- Math and Reasoning Workloads: Competition-style problems backed by a 97.3 AIME 2025 score
- Cost-Efficient Coding Agents: Software engineering tasks served by a 74.4% SWE-bench Verified model with only 11B active parameters
- Long-Context Analysis: Repository or document review that uses the full 262.1K tokens window
- High-Volume Production Traffic: Workloads where sparse MoE activation keeps per-request costs low
Consider Alternatives When
- Multimodal Inputs: StepFun 3.5 Flash is text-only, so route image or video requests to a vision-capable model
- Maximum Benchmark Headroom: Larger frontier models still lead on the hardest coding and research suites
- Simple Single-Turn Chat: Reasoning tokens add output overhead that lightweight conversation doesn't need
Conclusion
StepFun 3.5 Flash gives you open-weight frontier reasoning without dense-model inference costs. Strong AIME, SWE-bench, and Tau2-Bench results make StepFun 3.5 Flash a practical default for agents that think before they act. Route it through AI Gateway and you get failover, observability, and one key for every model in the catalog.