Gemma 3 27B is a 27-billion-parameter vision-language model from Google, handling text and image input and generating text output. It features a 128K context window, multilingual support, and open weights. Suitable for complex question answering, summarization, reasoning, and image understanding tasks.
发布日期2025年3月12日
参数规模27B
上下文长度40K
许可证Gemma
知识截止—
Benchmarks
评测成绩
| 评测基准 | 类别 | 分数 | 来源 |
|---|---|---|---|
| GSM8k | math reasoning | 95.9 | 来源 |
| 安全防御榜·Toxicity | security | 91.9 | — |
| IFEval | general | 90.4 | 来源 |
| 多模态安全榜·Bias | security vision | 90.1 | — |
| MATH | math reasoning | 89.0 | 来源 |
| HumanEval | reasoning code | 87.8 | 来源 |
| BIG-Bench Hard | reasoning math language | 87.6 | 来源 |
| DocVQA | vision multimodal | 86.6 | 来源 |
| 安全防御榜·Bias | security | 85.7 | — |
| AI2D | vision reasoning multimodal | 84.5 | 来源 |
| Natural2Code | reasoning general | 84.5 | 来源 |
| 多模态安全榜·Misinformation | security vision | 83.7 | — |
| 多模态安全榜·Health Risk | security vision | 82.7 | — |
| 安全防御榜·Info Safety | security | 82.3 | — |
| 安全防御榜 | security | 81.3 | — |
| 开源模型能力榜·IFEval | general | 81.0 | 来源 |
| 多模态安全榜 | security vision | 79.6 | — |
| 多模态安全榜·Legality | security vision | 79.4 | — |
| 安全防御榜·Human Rights | security | 79.0 | — |
| 安全防御榜·Malicious Use | security | 78.6 | — |
| 安全防御榜·Socioeconomic | security | 78.2 | — |
| ChartQA | reasoning vision multimodal | 78.0 | 来源 |
| 多模态安全榜·Toxicity | security vision | 77.8 | — |
| Global-MMLU-Lite | general language reasoning | 75.1 | 来源 |
| FACTS Grounding | reasoning | 74.9 | 来源 |
| MBPP | reasoning general | 74.4 | 来源 |
| 安全防御榜·Child Safety | security | 73.1 | — |
| VQAv2 (val) | vision multimodal language reasoning | 71.0 | 来源 |
| InfoVQA | vision multimodal | 70.6 | 来源 |
| 开源模型能力榜·MMLU-Pro | general | 67.8 | 来源 |
| MathVista-Mini | math vision multimodal | 67.6 | 来源 |
| MMLU-Pro | language reasoning math general | 67.5 | 来源 |
| TextVQA | vision multimodal image-to-text | 65.1 | 来源 |
| MMMU (val) | vision multimodal reasoning general | 64.9 | 来源 |
| 多模态安全榜·Privacy | security vision | 63.7 | — |
| 价值对齐榜·Fairness | security | 63.6 | — |
| HiddenMath | math reasoning | 60.3 | 来源 |
| Bird-SQL (dev) | reasoning | 54.4 | 来源 |
| WMT24++ | language | 53.4 | 来源 |
| 开源模型能力榜·GPQA-Diamond | general | 46.3 | 来源 |
| 价值对齐榜 | security | 44.9 | — |
| 价值对齐榜·Morality | security | 44.4 | — |
| 价值对齐榜·Data Protection | security | 44.4 | — |
| 价值对齐榜·Safety | security | 43.6 | — |
| GPQA | reasoning general | 42.4 | 来源 |
| 开源模型能力榜 | general | 42.1 | 来源 |
| 开源模型能力榜·LiveCodeBenchV6 | general | 30.8 | 来源 |
| LiveCodeBench | reasoning general code | 29.7 | 来源 |
| 价值对齐榜·Legality | security | 28.6 | — |
| 开源模型能力榜·AIME2025 | general | 22.4 | 来源 |
| BIG-Bench Extra Hard | reasoning general language | 19.3 | 来源 |
| ECLeKTic | language reasoning | 16.7 | 来源 |
| SimpleQA | general reasoning | 10.0 | 来源 |
| 开源模型能力榜·HLE | general | 4.2 | 来源 |
Pricing
API 价格对比
| 服务商 | 输入价 | 输出价 | 上下文 | 吞吐(tok/s) | 延迟(s) | 函数调用 | 代码执行 | 联网搜索 |
|---|---|---|---|---|---|---|---|---|
| Scaleway | $0.25 | $0.50 | 40K | — | — | ✓ | ✗ | ✗ |
价格单位：美元/百万 token，数据来自社区整理，仅供参考。