Mistral Large 2 vs Mistral Medium 3.5
Benchmarks, API Token Pricing, Context Window & Speed Comparison
Related topics (12)
Mistral AI 的 Mistral Large 2 与 Mistral AI 的 Mistral Medium 3.5 是当前备受关注的两款主力 AI 模型。Mistral Large 2 提供 128K Token 上下文窗口，API 定价为每百万 Token 输入 $2.00 / 输出 $6.00；而 Mistral Medium 3.5 提供 262.1K Token 上下文窗口，定价为每百万 Token 输入 $0.40 / 输出 $2.00。以下为两者的基准测试、延迟与成本全面对比。
切换并对比任意模型
Select any AI models from our verified benchmark database to generate live side-by-side matrices.
Mistral Large 2
开源权重基础大模型Mistral AI 推出的 Mistral Large 2 是一款支持私有云服务器自托管、微调与完全数据主权的开源权重基础大模型，具备 128K 上下文窗口、75 Token/秒生成速度与 46/100 综合智能指数（每百万 Token $2 输入 / $6 输出）。
$2.00 / 1M
$6.00 / 1M
128K tokens
38%
Mistral Medium 3.5
开源权重基础大模型Mistral AI 推出的 Mistral Medium 3.5 是一款支持私有云服务器自托管、微调与完全数据主权的开源权重基础大模型，具备 262K 上下文窗口、90 Token/秒生成速度与 55/100 综合智能指数（每百万 Token $0.4 输入 / $2 输出）。
$0.40 / 1M
$2.00 / 1M
262.1K tokens
54%
核心差异一览
- Mistral Medium 3.5 is 3.9x cheaper for API usage (input + output tokens combined).
- Mistral Medium 3.5 has a larger context window (262K vs 128K tokens).
- Mistral Medium 3.5 is faster at 90 tok/s vs 75 tok/s.
- Mistral Medium 3.5 leads on SWE-bench coding (54.0% vs 38.0%).
| Feature | Mistral Large 2 | Mistral Medium 3.5 |
|---|---|---|
Intelligence Index Composite reasoning & knowledge score (0-100) | 46 | 55 |
Provider | Mistral AI | Mistral AI |
Input Price Cost per million input tokens | $2.00 / 1M | $0.40 / 1M |
Output Price Cost per million output tokens | $6.00 / 1M | $2.00 / 1M |
Blended Price Weighted 7:2:1 cache:input:output per 1M | $1.70 / 1M | $0.35 / 1M |
Context Window Maximum input + output tokens | 128K tokens~192 A4 pages of text | 262.1K tokens~393 A4 pages of text |
Max Output Tokens Per single request | 8K | 33K |
Throughput Speed Tokens generated per second | 75 tok/s | 90 tok/s |
SWE-bench Verified Real-world software engineering | 38% | 54% |
GPQA Diamond Graduate-level reasoning | 54% | 68% |
Cached Input Price Prompt caching discount per 1M tokens | $1.00 / 1M | $0.10 / 1M |
各项指标谁更胜一筹？
At-a-glance decision framework权威基准测试得分对比
Official evaluation metrics verified by independent third-party research labs.
Industry benchmark evaluating real-world GitHub issue resolution and multi-file debugging.
Challenging questions created by domain experts to test complex PhD-level reasoning.
Hardened reasoning and academic knowledge benchmark with reduced lucky-guess probability.
Crowdsourced human preference evaluation based on blind, head-to-head model battles.
Autonomous multi-step tasks executed inside a real command-line terminal environment.
American Invitational Mathematics Examination — multi-step Olympiad-level reasoning.
Accuracy when reasoning over very large inputs that fill the model context window.
Tokens emitted per second during active generation streaming.
延迟与响应速度
Time to first token (TTFT) and estimated end-to-end response time for a typical 500-token response.
实时 Token 成本模拟器
Adjust sliders to calculate projected monthly cloud API spending with prompt caching discounts.
Mistral Large 2 · Strengths & Key Advantages
- 提供 128K 上下文窗口与最高 8K 输出容量，核心基准表现优异（SWE-bench Verified 38% · GPQA Diamond 54%）
- 高达 75 Token/秒的生成吞吐量，首字延迟 (TTFT) 仅约 520 毫秒
- 开放模型权重 (123B)，支持通过 vLLM、SGLang 或 Ollama 在私有 GPU 集群上本地部署与微调
Considerations & Trade-offs
- 在本地以满血精度运行需要较高的 GPU 显存 (VRAM) 与运维基础设施
- 在大规模高频调用场景下，建议结合提示词缓存 (Prompt Caching) 与批量 API 以控制成本
Mistral Medium 3.5 · Strengths & Key Advantages
- 提供 262K 上下文窗口与最高 33K 输出容量，核心基准表现优异（SWE-bench Verified 54% · GPQA Diamond 68%）
- 高达 90 Token/秒的生成吞吐量，首字延迟 (TTFT) 仅约 450 毫秒
- 开放模型权重 (Confidential (~80B MoE))，支持通过 vLLM、SGLang 或 Ollama 在私有 GPU 集群上本地部署与微调
Considerations & Trade-offs
- 在本地以满血精度运行需要较高的 GPU 显存 (VRAM) 与运维基础设施
- 在超长链路的前沿科研推导任务上，深度略逊于顶配重型旗舰推理模型
Token Use & Task Economics
Estimated cost and token consumption to complete one representative agentic task.
| Metric | Mistral Large 2 | Mistral Medium 3.5 |
|---|---|---|
Cost per Task Estimated USD to finish one task | $0.20 | $0.38 |
Output Tokens / Task Tokens generated per task | 24K | 38K |
Reasoning Tokens / Task Thinking tokens burned per task | 0K | 18K |
技术规格参数
Architecture, release timeline, capabilities, and modalities.
| Feature | Mistral Large 2 | Mistral Medium 3.5 |
|---|---|---|
| Official API Model ID (SDK) | mistral-large-2-latest | mistral-medium-3-5-latest |
| API Protocol | OpenAI-Compatible REST / vLLM / SGLang / HuggingFace TGI | OpenAI-Compatible REST / vLLM / SGLang / HuggingFace TGI |
| Release Date | 2024-07 | 2026-04 |
| Knowledge Cutoff | July 2024 | February 2026 |
| Architecture | 123B Dense Frontier Transformer | Enterprise Hybrid Reasoning MoE |
| Parameters | 123B | Confidential (~80B MoE) |
| Enterprise Rate Limits (RPM / TPM) | Unlimited (Hardware-bound on self-hosted GPU) / 10,000 RPM Cloud · Unlimited local throughput / 10M+ TPM Managed Cloud | Unlimited (Hardware-bound on self-hosted GPU) / 10,000 RPM Cloud · Unlimited local throughput / 10M+ TPM Managed Cloud |
| Long-Context Tier (>128K/200K) | Flat rate ($2/1M in) across full 128K window | Flat rate ($0.4/1M in) across full 262K window |
| Prompt Cache Policy & TTL | Cached input at $1/1M tokens | Cached input at $0.1/1M tokens |
| Computer Use / GUI Agent | API / CLI Tool-Use & DOM Agent ready (no raw pixel desktop click loop) | API / CLI Tool-Use & DOM Agent ready (no raw pixel desktop click loop) |
| MCP & Parallel Tool Calling | Native MCP Client/Server ready · Parallel function calling · 100% Strict JSON Schema | Native MCP Client/Server ready · Parallel function calling · 100% Strict JSON Schema |
| Web Search Grounding & Sandbox | Supported via external Tavily/Exa Search & E2B/Docker Sandbox tools | Supported via external Tavily/Exa Search & E2B/Docker Sandbox tools |
| Official Fine-Tuning Support | Full Weights SFT, LoRA, QLoRA, DPO & GRPO RL (Unsloth / Axolotl / Llama-Factory) | Full Weights SFT, LoRA, QLoRA, DPO & GRPO RL (Unsloth / Axolotl / Llama-Factory) |
| Multimodal Token Billing | Text BPE (~4 chars/token) | Text BPE (~4 chars/token) · Vision (~258–765 tokens per image tile) |
| Quantization & Min GPU VRAM | BF16 / FP16, FP8 Dynamic, INT4 AWQ / GPTQ, GGUF (Q4_K_M / Q8_0) · 76 GB VRAM (4-bit Quantized) · 264 GB VRAM (Native BF16) | BF16 / FP16, FP8 Dynamic, INT4 AWQ / GPTQ, GGUF (Q4_K_M / Q8_0) · 50 GB VRAM (4-bit Quantized) · 172 GB VRAM (Native BF16) |
| Safety & Frontier Classification | Open-Weights System Card + Customizable Guardrail Layer (Llama Guard / ShieldGemma) | Open-Weights System Card + Customizable Guardrail Layer (Llama Guard / ShieldGemma) |
| Enterprise Compliance (ZDR/HIPAA/SOC2) | 100% Air-Gapped On-Premises / Private VPC · Full HIPAA, GDPR & SOC2 Control | 100% Air-Gapped On-Premises / Private VPC · Full HIPAA, GDPR & SOC2 Control |
| Input Modalities | text | text, vision |
| Output Modalities | text | text |
Mistral Large 2 vs Mistral Medium 3.5: 常见问题解答 (FAQ)
Which is more intelligent, Mistral Large 2 or Mistral Medium 3.5?
Mistral Medium 3.5 is more intelligent. It scores 55 compared with 46 on the composite Intelligence Index.
Which model is faster?
Mistral Medium 3.5 is faster, generating 90 tokens per second compared with 75 tokens per second.
Which model is cheaper to run?
Mistral Medium 3.5 is cheaper at $0.35 per 1M tokens blended, compared with $1.70 (7:2:1 cache:input:output ratio).
Which model has lower latency?
Mistral Medium 3.5 has lower latency, with a time to first token of 0.45s compared with 0.52s.
Which model has a larger context window?
Mistral Medium 3.5 has a larger context window, supporting 262.1K tokens compared with 128K tokens.
综合结论
综合来看，Mistral Large 2（适合：支持私有云服务器自托管、微调与完全数据主权的开源权重基础任务、企业级 RAG 检索增强生成、自动化智能体工作流与高可靠 API 集成。）与 Mistral Medium 3.5（适合：支持私有云服务器自托管、微调与完全数据主权的开源权重基础任务、企业级 RAG 检索增强生成、自动化智能体工作流与高可靠 API 集成。）各具优势：Mistral Medium 3.5 在综合智能与编程基准测试中表现更强，Mistral Medium 3.5 拥有更快的 Token 生成速度，而 Mistral Medium 3.5 则在规模化 API 部署中具备更高的成本效益。
常见问题 (FAQ)
在编程和深度推理方面，Mistral Large 2 和 Mistral Medium 3.5 哪个更强？+
在综合智能指数与编程基准评测中，Mistral Medium 3.5 表现占优。建议结合上方的 SWE-bench Verified 与 GPQA Diamond 跑分根据具体场景选择。
Mistral Large 2 与 Mistral Medium 3.5 的 API 价格哪个更便宜？+
Mistral Large 2 的 API 价格为输入 $2.00 / 输出 $6.00（每百万 Token），Mistral Medium 3.5 为输入 $0.40 / 输出 $2.00。综合混合成本来看，Mistral Medium 3.5 更具性价比。
哪款模型拥有更大的上下文窗口和更快的响应速度？+
Mistral Large 2 支持 128K Token 上下文窗口，而 Mistral Medium 3.5 支持 262.1K Token。在流式生成速度方面，Mistral Medium 3.5 的吞吐表现更快。
Understand key metrics like context window, token pricing, reasoning mode, and SWE-bench scores in our comprehensive AI glossary.
Related reading
- Mistral Large 2 vs GPT-4o：基准测试、API 价格与速度全面对比 (2026)对比
- Mistral Large 2 vs Llama 3.3 70B：基准测试、API 价格与速度全面对比 (2026)对比
- Mistral Medium 3.5 vs Claude Sonnet 5.5：基准测试、API 价格与速度全面对比 (2026)对比
- Mistral Medium 3.5 vs DeepSeek V4.1-Flash：基准测试、API 价格与速度全面对比 (2026)对比
- Mistral Medium 3.5 vs GPT-5.5：基准测试、API 价格与速度全面对比 (2026)对比
- Pixtral 12B vs Mistral Large 2：基准测试、API 价格与速度全面对比 (2026)对比
Benchmark scores, token prices, and rate limits evolve rapidly as AI labs release model updates. Always verify current specifications and pricing on the official provider documentation.
数据来源: Mistral AI 与 Mistral AI 官方 API 文档、模型发布说明及独立基准测试报告 (2026)。