来自 DeepMind Gemini 3.7 Flash 模型卡表的值。 · 来源
来自 DeepMind Gemini 3.7 Flash 模型卡表的值。 · 来源
来自 DeepMind Gemini 3.7 Flash 模型卡表的值。 · 来源
| 模型 | 输入$/1M | 产出 $/100 万美元 | 人工智能分析智能指数 |
|---|---|---|---|
| 双子座3.7闪存 | 0.75 美元(介绍性) | 3.75 美元(介绍性) | 56 |
| 双子座3.6闪存 | 0.75 美元(介绍性) | 3.75 美元(介绍性) | 52 |
| 克劳德十四行诗 5 | $2.00 | $10.00 | 55 |
| GPT-5.6 泰拉 | $2.00 | $12.00 | 57 |
| 缪斯火花1.2 | $1.25 | $4.25 | 57 |
如 DeepMind 的模型卡中所列;Google 的价格是截至 2026 年底的介绍性价格。 · 来源
Estimate from the version's published specs, benchmarks and reviews, 0-5; scores are only comparable within one criteria year. Scored as launched (Gemini 3.7 Flash, 13 Aug 2026): Artificial Analysis Intelligence Index 56 at high reasoning (AA article of 13 Aug; its model page now shows 39 on v4.3.2, not used); 1M-token context; launch list price was the introductory $0.75 in / $3.75 out per 1M through end of 2026 = $1.50 blended at 3:1 (scored 4.5); the standard $1.50/$7.50 = $3.00 blended would score 4; about 340 output tokens/s per AA; text, image, video, speech input. Coding null: no Terminal-Bench 4.0 at launch (Google 85.8% on Terminal-bench 2.1; 14.9% on 3.0). Reviews: AA's note, a Bijan Bowen hands-on video (summary page), buildfastwithai; one hands-on source only.[1]
Gemini 3.7 Flash (high) scores 56 on the Artificial Analysis Intelligence Index, just behind GPT-5.6 Terra (max, 57) and Muse Spark 1.2 (xhigh, 57)
Gemini 3.7 Flash is fast and comparatively inexpensive, but its hands-on coding results improve on its predecessor without reaching uniformly reliable frontier performance.
Gemini 3.7 Flash is a genuine upgrade over 3.6 Flash, with large gains on coding and agent benchmarks like DeepSWE
读者上传的照片,与图钉带来源的媒体分开显示。你可以为吃到或看到的内容评分,在餐厅图钉上还能直接在照片上标记菜品。
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。
9to5Google的发布报道日期为2026年8月13日,称谷歌“今天宣布了Gemini 3.7 Flash”;DeepMind[2]的模型卡于2026年8月13日发布。
DeepMind's model card, published 13 August 2026, lists the introductory prices and a benchmark table against 3.6 Flash, Claude Sonnet 5, GPT-5.6 Terra and Muse Spark 1.2.
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: Gemini 3.7 Flash Released
| 基准测试 | Gemini 3.7 Flash | Gemini 3.6 Flash | Claude Sonnet 5 | GPT-5.6 Terra | Muse Spark 1.2 |
|---|---|---|---|---|---|
| 输入价格(美元/百万token) | $0.75* | $0.75* | $2.00 | $2.00 | $1.25 |
| 输出价格(美元/百万token) | $3.75* | $3.75* | $10.00 | $12.00 | $4.25 |
| Artificial Analysis Intelligence Index(综合模型智能) | 56 | 52 | 55 | 57 | 57 |
| FrontierCode 1.1 Main(生产级代码质量,得分) | 43.6% | 34.4% | 42.7% | 41.3% | — |
| DeepSWE v1.1(长周期软件工程) | 65.3% | 48.6% | 53.8% | 69.6% | 54.9% |
| Code Arena(网页开发,Elo) | 1588 | 1538 | 1541 | 1523 | 1535 |
| Terminal-bench 2.1(智能体终端编程) | 85.8% | 78.0% | 80.4% | 87.4% | 82.9% |
| Terminal-bench 3.0(通用智能体能力) | 14.9% | 5.4% | 14.6% | 20.8% | — |
| AutomationBench(企业工作流自动化,私有测试集) | 30.4% | 17.0% | 10.7% | 23.6% | — |
| GDPVal-AA v2(知识工作,Elo) | 1525 | 1422 | 1598 | 1578 | 1628 |
| Harvey LAB-AA(复杂法律工作流) | 90.7% | 85.1% | 90.1% | 85.2% | — |
| GDP.pdf(专家级PDF文档理解) | 34.0% | 22.0% | 28.0% | 24.7% | 16.0% |
| CharXiv Reasoning(复杂图表的信息综合,不使用工具) | 84.5% | 85.2% | 77.0% | 85.9% | — |
| CharXiv Reasoning(复杂图表的信息综合,使用工具) | 88.7% | 89.4% | 88.3% | — | — |
| LVBench(长视频理解) | 85.4% | 84.2% | 68.5% | 78.9% | — |
| GDM-MRCR v2 (8-needle)(长上下文表现,128k(平均)) | 97.0% | 91.8% | 81.5% | 93.5% | — |
| OSWorld-2.0(智能体计算机操作) | 47.9% | 33.8% | — | 50.2% | — |
| Agent's Last Exam(多模态桌面和操作系统智能体任务,通过率) | 26.3% | 24.2% | 33.3% | 28.0% | — |
| HLE-Verified(多学科专家推理) | 53.6% | 51.2% | 31.0% | 51.1% | — |
| BioMysteryBench(生物信息学研究推理,人类可解) | 87.1% | 80.6% | 87.5% | 83.8% | — |
| BioMysteryBench(生物信息学研究推理,人类难解) | 43.5% | 41.2% | 34.1% | 49.4% | — |
| LABBench2(生物学真实科研任务) | 82.1% | 76.1% | 80.1% | 81.2% | — |