这些是读者照片,与图钉带来源的媒体分开显示。
第一项始终是图钉的来源。总体可信度是各份资料对上方所用开始和结束时间支持程度的加权平均;资料每比最新的一份旧 180 天,权重减半。
AIBase[6] 在 2026 年 2 月 2 日的报道称其为 StepFun[3]“最新的开源基础模型”,代码在 GitHub,权重在 Hugging Face[2],GitHub 仓库的首次提交日期为 2 月 2 日;OpenRouter 自 1 月 29 日起上架,Hugging Face 仓库于 2 月 1 日(UTC)创建,因此公开发布日为 2 月 2 日。
The model card: Apache 2.0, 196B total and about 11B active parameters with 288 routed experts plus one shared and top-8 routing, 256K context, 100-300 tok/s typical and about 350 tok/s peak, 74.4% SWE-bench Verified and 51.0% Terminal-Bench 2.0, recommended sampling.[1]
The API docs: step-3.5-flash at $0.10 per 1 million input tokens, $0.02 cache hit and $0.30 output; a step-3.5-flash-2603 agent-optimized variant with a low-reasoning mode; OpenAI-compatible chat completions with tool calling.[1]
ifanr (6 March): StepFun[1][3] fully open-sourced the model in early March, releasing the Base and Midtrain weights and the Steptron training framework alongside it, with Step 3.5 Flash's OpenClaw call volume reaching first place globally.
The technical report (216 authors, submitted 11 February, revised 23 February): 3:1 sliding-window/full attention, MTP-3, a reinforcement learning framework, and 85.4% IMOAnswerBench, 86.4% LiveCodeBench-v6, 88.2% tau2-Bench, 69.0% BrowseComp (with context management).[1]
有遗漏或错误吗?用你自己的话说明:能佐证此图钉的链接、不同的开始或结束日期及理由,或缺失、有误的信息。AI 会对照此图钉的来源进行核实,搜索更好的来源,并添加任何支持你说法的页面。图钉自身的来源仍然最重要。AI 也会查看图片:显示的是别的东西或显示效果差的图片会被移到后面或替换。
原标题: Step 3.5 Flash Released
| 基准 | Step 3.5 Flash | DeepSeek V3.2 | Kimi K2 Thinking / K2.5 | GLM-4.7 | MiniMax M2.1 | MiMo-V2 Flash |
|---|---|---|---|---|---|---|
| 激活 / 总参数量 | 11B / 196B | 37B / 671B | 32B / 1T | 32B / 355B | 10B / 230B | 15B / 309B |
| 128K 下的预估解码成本(Hopper) | 1.0x | 6.0x | 18.9x | 18.9x | 3.9x | 1.2x |
| τ²-Bench | 88.2 | 80.3 (85.2*) | 74.3* / 85.4* | 87.4 | 86.6* | 80.3 (84.1*) |
| BrowseComp | 51.6 | 51.4 | 41.5* / 60.6 | 52.0 | 47.4 | 45.4 |
| BrowseComp (使用 Context Manager) | 69.0 | 67.6 | 60.2 / 74.9 | 67.5 | 62.0 | 58.3 |
| GAIA (无文件) | 84.5 | 75.1* | 75.6* / 75.9* | 61.9* | 64.3* | 78.2* |
| AIME 2025 | 97.3 | 93.1 | 94.5 / 96.1 | 95.7 | 83.0 | 94.1 (95.1*) |
| HMMT 2025 (2月) | 98.4 | 92.5 | 89.4 / 95.4 | 97.1 | 71.0* | 84.4 (95.4*) |
| IMOAnswerBench | 85.4 | 78.3 | 78.6 / 81.8 | 82.0 | 60.4* | 80.9* |
| LiveCodeBench-V6 | 86.4 | 83.3 | 83.1 / 85.0 | 84.9 | — | 80.6 (81.6*) |
| SWE-bench Verified | 74.4 | 73.1 | 71.3 / 76.8 | 73.8 | 74.0 | 73.4 |
| Terminal-Bench 2.0 | 51.0 | 46.4 | 35.7* / 50.8 | 41.0 | 47.9 | 38.5 |