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。
9月20日的RuntimeWire[6]报道称StepFun[2]当天“launches”该模型,《每日经济新闻》称StepFun于9月20日(9月20日)发布;文档页面本身没有日期。维基百科的9月18日与OpenRouter 10月8日的路由列表相互矛盾,均不是发布日。
StepFun's[1] pricing table: step-5-preview at $1.00 per 1 million input tokens (cache miss), $0.10 cache hit and $2.70 output, output including the reasoning trace; step-3.7-flash $0.20, $0.04 and $1.15; step-3.5-flash $0.10, $0.02 and $0.30.
Independent evaluation: 44 on the Intelligence Index v4.3.2 (40th of 218), 88.5 output tokens per second, $1.03 per Intelligence Index task, 160M output tokens in the index run (very verbose), 1M context, text and image input, listed as a proprietary model released in September 2026.[1]
Daily Economic News on 21 September: StepFun[1][2] released Step 5 Preview on 20 September, 600B total and 27B active, 1M context, 44 on the Artificial Analysis[3] index (Qwen3.8 Max and GLM-5.3 score 45), DeepSWE v1.1 67.7%, single-task cost 1/8 of Claude Opus 5, and an analyst calling it a "textbook" pass that is one step short of a generational leap.
Third-party write-up that reproduces StepFun's[1][2] launch charts and tables: the benchmark comparison against Kimi K3, GLM-5.3, GPT-6 Astra and Claude Opus 5, 64k max output, up to 60 images per request, the 15 October open-weights countdown and the -65% cost Pareto chart.
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 5 Preview Released
| 基准测试(阶跃星辰公布) | Step 5 Preview | Kimi K3 | GLM-5.3 | GPT-6 Astra | Claude Opus 5 |
|---|---|---|---|---|---|
| DeepSWE v1.1 | 67.7 | 67.5 | 66.9 | 74.1 | 74.0 |
| StepCodeBench(阶跃星辰自研) | 49.0 | 43.9 | 40.2 | 61.0 | 63.9 |
| ProgramBench | 80.5 | 77.8 | 72.0 | 85.4 | 82.3 |
| Terminal-Bench v4 | 33.3 | 12.6 | 41.9 | 57.9 | 52.3 |
| FrontierFinance | 66.4 | 62.6 | 64.1 | 55.0 | 69.7 |