StepFun 發布 Step 3.7 Flash：一個用於編碼代理和搜索工作流程的 198B MoE 視覺語言模型

StepFun today released Step 3.7 Flash, a multimodal Mixture-of-Experts model targeting agentic use cases. It adds native vision input and improved tool-use reliability over Step 3.5 Flash. What is Step 3.7 Flash? Step 3.7 Flash is a 198B-parameter sparse Mixture-of-Experts (MoE) vision-language model. It pairs a 196B-parameter language backbone with a 1.8B-parameter vision encoder (ViT) for native image understanding. The model activates approximately 11B parameters per token during inference. In MoE architectures, only a subset of “expert” sub-networks fires per forward pass — not the full network. This keeps inference compute closer to an 11B dense model while maintaining a 198B total parameter budget. Key specs: SpecValueTotal parameters198B (196B language + 1.8B ViT)Active parameters per token~11BContext window256k tokensThroughputUp to 400 tokens/secReasoning levelsLow, medium, highLicenseApache 2.0 Architecture Notes The vision encoder runs as a separate 1.8B ViT module. It injects image representations into the language backbone’s context. Step 3.5 Flash had no multimodal support; this is a new addition in 3.7. Three selectable reasoning depths — low, medium, and high — let developers trade latency for reasoning depth. Low is faster and cheaper; high applies more computation per response. Agentic Coding Performance On SWE-Bench Pro, Step 3.7 Flash scores 56.26%, up from Step 3.5 Flash’s 51.3% — a gain of roughly 5 percentage points. On Terminal-Bench 2.1, it scores 59.55%, up from 53.37%. On SWE-MTLG (a multi-task long-generation coding benchmark), it scores 72.42%. Cross-harness consistency on StepFun’s internal Step-SWE-Bench: ScaffoldStep 3.7 FlashStep 3.5 FlashHermes Agent67.5%60.0%OpenClaw67.0%47.0%KiloCode67.5%59.0%RooCode64.5%43.0%Claude Code71.5%73.0%OpenCode64.5%57.0% Step 3.5 Flash ranged from 43% to 73% across harnesses. Step 3.7 Flash ranges from 64.5% to 71.5%. In production, coding agents often run inside heterogeneous scaffolds — each with its own prompting conventions and tool schemas. Narrower per-harness variance means more predictable behavior across different setups. Advisor Mode Step 3.7 Flash supports Advisor Mode, StepFun’s implementation of the advisor strategy described by Anthropic. The model runs the agentic loop end-to-end — calling tools, reading results, iterating — and escalates to a larger advisor model only at specific inflection points, such as planning or recovering from repeated failures. Most of the run stays at executor cost. With Advisor Mode enabled on SWE-Bench Verified, StepFun reports Step 3.7 Flash reaches 97% of Claude Opus 4.6’s coding performance at roughly one-ninth the per-task cost ($0.19 vs. $1.76 per task). These are StepFun’s internal figures. Multimodal Capabilities Step 3.7 Flash supports two visual tool pathways: Visual Search Tool — For recognition tasks where the model’s parametric knowledge is insufficient (long-tail entities, recently emerged concepts), it invokes a visual search tool to retrieve and verify. On SimpleVQA (with Search), it scores 79.16%, comparable to GPT 5.5 (79.11%) and above Kimi K2.6 (78.24%) and GLM 5V Turbo (78.20%). Python Tool — For fine-grained visual tasks (high-resolution images, visual probing, bounding-box analysis), it uses a code interface to crop, zoom, and draw pixels or bounding boxes. On V (a self-tested score with Python), it scores 95.29%. On HR-Bench 4K and HR-Bench 8K, it scores 89.13% and 86.34% respectively. StepFun notes an observed behavior during testing: the model combined visual tools with non-visual tools without being explicitly trained to do so. For example, after generating frontend code, it used the GUI to render and inspect the result before iterating. StepFun describes this as emergent compositional tool use. On Android Daily (long-horizon phone UI task completion), Step 3.7 Flash scores 61.87%, ahead of Kimi K2.6 (53.36%) and GLM 5V Turbo (51.68%). Gemini 3 Flash (63.21%) leads this benchmark. Search and Research Benchmarks StepFun focused this model’s search design on planning, evidence filtering, and synthesis — integrating search as part of the reasoning loop rather than a separate add-on. BenchmarkStep 3.7 FlashNotable comparisonHLE with Tools (acc)47.20%DeepSeek V4 Flash: 45.10%BrowseComp (acc)75.82%Claude Opus 4.7: 79.30%DeepSearchQA (F1)92.82%Kimi K2.6: 92.50%ResearchRubrics (score)71.68%GPT 5.5: 61.50% Note: The HLE with Tools score of 47.20% compares to Step 3.5 Flash’s text-only score of 35.68%. Step 3.5 Flash did not support tool-augmented evaluation on HLE. General Agent Benchmarks BenchmarkStep 3.7 FlashDescriptionToolathlon49.51%Multi-tool coordinationClawEval-1.167.07%Daily autonomous task execution in realistic environmentsGDPval (44 occupations)45.8%General professional task executionTau2-bench Telecom>98%Across different reasoning difficulty tiers On ClawEval-1.1, Step 3.7 Flash (67.07%) leads DeepSeek V4 Flash (57.80%) and DeepSeek V4 Pro (59.80%) among the compared models. Long-Context Performance On AA-LCR (a long-context retrieval benchmark, avg@16/acc), Step 3.7 Flash scores 63.94%. This is comparable to DeepSeek V4 Flash (63.70%) and DeepSeek V4 Pro (66.30%). Pricing Token TypePriceInput (cache miss)$0.20 / M tokensInput (cache hit)$0.04 / M tokensOutput$1.15 / M tokens Marktechpost’s Visual Explainer #s37f-guide*{box-sizing:border-box;margin:0;padding:0} #s37f-guide{font-family:'Segoe UI',system-ui,sans-serif;background:#f7f8fa!important;border:1.5px solid #e2e6ed!important;border-radius:14px!important;overflow:hidden!important;max-width:720px!important;margin:0 auto!important;box-shadow:0 4px 24px rgba(0,0,0,.07)!important} #s37f-guide .sf-header{background:#ffffff!important;border-bottom:1.5px solid #e2e6ed!important;padding:18px 24px 14px!important} #s37f-guide .sf-tag{display:inline-block!important;background:#eef2ff!important;color:#4361ee!important;font-size:10px!important;font-weight:700!important;letter-spacing:.08em!important;text-transform:uppercase!important;padding:3px 10px!important;border-radius:20px!important;margin-bottom:8px!important} #s37f-guide .sf-title{font-size:17px!important;font-weight:700!important;color:#111827!important;line-height:1.35!important} #s37f-guide .sf-sub{font-size:12px!important;color:#6b7280!important;margin-top:4px!important} #s37f-guide .sf-progress{height:3px!important;background:#e9ecf0!important;position:relative!important} #s37f-guide .sf-progress-bar{height:3px!important;background:#4361ee!important;transition:width .35s ease!important;position:absolute!important;top:0!important;left:0!important} #s37f-guide .sf-slides{position:relative!important;overflow:hidden!important;height:560px!important} #s37f-guide .sf-slide{display:none!important;padding:24px!important;height:560px!important;overflow-y:auto!important;background:#f7f8fa!important;animation:sfIn .25s ease!important} #s37f-guide .sf-slide.active{display:block!important} @keyframes sfIn{from{opacity:0;transform:translateY(6px)}to{opacity:1;transform:translateY(0)}} #s37f-guide .sf-slide-label{font-size:10px!important;font-weight:700!important;letter-spacing:.1em!important;text-transform:uppercase!important;color:#4361ee!important;margin-bottom:10px!important} #s37f-guide .sf-slide h3{font-size:15px!important;font-weight:700!important;color:#111827!important;margin-bottom:12px!important;line-height:1.4!important} #s37f-guide .sf-slide p{font-size:12.5px!important;color:#374151!important;line-height:1.6!important;margin-bottom:8px!important} #s37f-guide .sf-slide p:last-child{margin-bottom:0!important} #s37f-guide .sf-kv{display:grid!important;grid-template-columns:1fr 1fr!important;gap:8px!important;margin-top:12px!important} #s37f-guide .sf-kv-item{background:#ffffff!important;border:1px solid #e2e6ed!important;border-radius:8px!important;padding:10px 12px!important} #s37f-guide .sf-kv-item .k{font-size:10px!important;color:#6b7280!important;font-weight:600!important;text-transform:uppercase!important