谷歌 DeepMind 模型卡。 · 来源
谷歌 DeepMind 模型卡。 · 来源
Google DeepMind 模型卡(UI:1/1/2025-5/1/2025 窗口)。 · 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 (3 Mar 2026, preview). Reasoning: Artificial Analysis Intelligence Index 34 at launch (v4.0; The Decoder 3 Mar 2026), nearest anchor 35. Coding null: no Terminal-Bench 4.0 (Google published LiveCodeBench 72.0%). Context 1M tokens (DeepMind model card). Price $0.25 in / $1.50 out per 1M = $0.5625 blended at 3:1. Speed 363 output tokens/s (Google's benchmark table, Artificial Analysis data; The Decoder: over 360). Input text, image, audio, video.[1]
On multimodal tasks, it beats top-tier models like Claude Opus 4.6 and Kimi K2.5, hitting 78 percent on the MMMU-Pro benchmark.
While Claude also has the same 1-million token context window, this model is arguably the best model for summarizing boring paperwork because it's so cheap to run.
The results aren't consistent, and the content quickly drifts into nonsense, but with tight guardrails, there could be some interesting use cases, like quick UI mockups to visualize ideas.
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。
谷歌文章《Gemini 3.1 Flash-Lite: Built for intelligence at scale》日期为2026年3月3日,并称当天起向开发者通过Gemini API、向企业通过Vertex AI推出预览版。
The Wikipedia article lists Gemini 3.1 Flash-Lite as released on 3 March 2026.
DeepMind's model card gives the benchmark table against the model's predecessor and named rivals.
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 3.1 Flash-Lite Released
| 基准测试 | Gemini 3.1 Flash-Lite | Gemini 2.5 Flash | Gemini 2.5 Flash-Lite | GPT-5 mini | Claude 4.5 Haiku | Grok 4.1 Fast |
|---|---|---|---|---|---|---|
| 输入价格(美元/百万token,无缓存,越低越好) | $0.25 | $0.30 | $0.10 | $0.25 | $1.00 | $0.20 |
| 输出价格(美元/百万token,越低越好) | $1.50 | $2.50 | $0.40 | $2.00 | $5.00 | $0.50 |
| 输出速度(token/秒) | 363 | 249 | 366 | 71 | 108 | 145 |
| Humanity’s Last Exam(学术推理,完整集,文本+多模态,不使用工具) | 16.0% | 11.0% | 6.9% | 16.7% | 9.7% | 17.6% |
| GPQA Diamond(科学知识,不使用工具) | 86.9% | 82.8% | 66.7% | 82.3% | 73.0% | 84.3% |
| MMMU-Pro(多模态理解与推理,不使用工具) | 76.8% | 66.7% | 51.0% | 74.1% | 58.0% | 63.0% |
| CharXiv Reasoning(复杂图表的信息综合) | 73.2% | 63.7% | 55.5% | 75.5% | 61.7% | 31.6% |
| Video-MMMU(从视频中获取知识) | 84.8% | 79.2% | 60.7% | 82.5% | — | 74.6% |
| SimpleQA Verified(参数化知识) | 43.3% | 28.1% | 11.5% | 9.5% | 5.5% | 19.5% |
| FACTS Benchmark Suite(涵盖依据、参数化、搜索和多模态的事实性基准) | 40.6% | 50.4% | 17.9% | 33.7% | 18.6% | 42.1% |
| MMMLU(多语言问答) | 88.9% | 86.6% | 84.5% | 84.9% | 83.0% | 86.8% |
| LiveCodeBench(代码生成(UI:2025年1月1日至5月1日)) | 72.0% | 62.6% | 34.3% | 80.4% | 53.2% | 76.5% |
| MRCR v2 (8-needle)(长上下文表现,128k(平均)) | 60.1% | 54.3% | 30.6% | 52.5% | 35.3% | 54.6% |
| MRCR v2 (8-needle)(长上下文表现,1M(逐点)) | 12.3% | 21.0% | 5.4% | — | — | 6.1% |