返回开发者专区官方定价 官方定价 官方定价
Mistral 模型
探索 Mistral 的所有 3 个模型，包括详细定价、优缺点和开发者推荐。
3
模型
$0.400
最低输入价格
131K
最大上下文
2
质量层级
快速推荐
最佳性价比: Mistral Medium 3 ($0.400/1M)
最佳质量: Mistral Large 2
Mistral Large 2
FlagshipMultilingual, complex tasks
适用场景: Top pick for multilingual enterprise apps, especially EU-based deployments.
核心升级
- ◆MMLU: 84.0% — competitive with GPT-4 class models
- ◆Fine-tuning available — customize for domain-specific tasks
- ◆EU-based (GDPR compliant) — only major flagship with EU sovereignty
- ◆12 languages optimized — best-in-class for French, German, Spanish
- ◆128K context is smaller vs 1M competitors — trade-off for EU compliance
输入价格
$2.00
per 1M tokens
输出价格
$6.00
per 1M tokens
缓存输入
—
per 1M tokens
批量输入
—
per 1M tokens
上下文窗口: 131K
最大输出: 8,192 tokens
知识截止日期: 2024-07
视觉函数调用微调JSON 模式免费层级
优点
- Excellent multilingual (French, German, etc.)
- Fine-tuning available
- EU-based (GDPR friendly)
缺点
- No vision
- 128K context is small vs competitors
- No cached/batch pricing
性能
输出速度~55 tok/s
速率限制3,000 RPM
多模态能力
图像输入图像输出音频输入音频输出
基准测试
MMLU
84.0%
HumanEval
82.0%
MATH
65.0%
Mistral Medium 3
Mid-tierBalanced multilingual
适用场景: Cost-effective multilingual + vision for European market applications.
核心升级
- ◆Vision added at mid-tier — multimodal capability at $0.40/M input
- ◆5x cheaper than Mistral Large ($0.40 vs $2/M) for lighter tasks
- ◆EU-based infrastructure — GDPR compliant for European data
- ◆128K context — sufficient for most enterprise document processing
- ◆No fine-tuning — upgrade to Large 2 for custom model training
输入价格
$0.400
per 1M tokens
输出价格
$2.00
per 1M tokens
缓存输入
—
per 1M tokens
批量输入
—
per 1M tokens
上下文窗口: 131K
最大输出: 8,192 tokens
知识截止日期: 2024-07
视觉函数调用微调JSON 模式免费层级
优点
- Vision at mid-tier price
- Good multilingual balance
- EU-based
缺点
- No fine-tuning
- No cached/batch pricing
- Smaller context than competitors
性能
输出速度~75 tok/s
速率限制5,000 RPM
多模态能力
图像输入图像输出音频输入音频输出
基准测试
MMLU
80.5%
HumanEval
78.0%
Mixtral 8x22B
Mid-tierOpen-weight, high throughput
适用场景: For self-hosting or fine-tuning on domain-specific data with high throughput needs.
核心升级
- ◆Open-weight MoE: 8x22B params, only 39B active per token — high throughput
- ◆Fine-tunable — full weight access for domain adaptation
- ◆Function calling + JSON mode — enterprise-ready tool integration
- ◆65K context — smaller than newer models but sufficient for most tasks
- ◆Self-hostable — no per-token cost when running on own infrastructure
输入价格
$0.900
per 1M tokens
输出价格
$2.70
per 1M tokens
缓存输入
—
per 1M tokens
批量输入
—
per 1M tokens
上下文窗口: 66K
最大输出: 4,096 tokens
知识截止日期: 2024-01
视觉函数调用微调JSON 模式免费层级
优点
- Open-weight MoE architecture
- Fine-tunable
- High throughput via sparse activation
缺点
- Only 65K context
- No vision
- Older knowledge cutoff (2024-01)
性能
输出速度~85 tok/s
速率限制—
多模态能力
图像输入图像输出音频输入音频输出
基准测试
MMLU
77.8%
HumanEval
75.5%
MATH
58.0%
并排比较
| 模型 | 层级 | 输入 | 输出 | 上下文 |
|---|---|---|---|---|
| Mistral Large 2 | Flagship | $2.00 | $6.00 | 131K |
| Mistral Medium 3 | Mid-tier | $0.400 | $2.00 | 131K |
| Mixtral 8x22B | Mid-tier | $0.900 | $2.70 | 66K |