GPT-5.2 Codex vs Devstral 2
Benchmarks, API Token Pricing, Context Window & Speed Comparison
Related topics (12)
OpenAI 的 GPT-5.2 Codex 与 Mistral AI 的 Devstral 2 是当前备受关注的两款主力 AI 模型。GPT-5.2 Codex 提供 400K Token 上下文窗口，API 定价为每百万 Token 输入 $1.75 / 输出 $14.00；而 Devstral 2 提供 262.1K Token 上下文窗口，定价为每百万 Token 输入 $1.50 / 输出 $4.50。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.
GPT-5.2 Codex
自主智能体与工具专家OpenAI 推出的 GPT-5.2 Codex 是一款专为长周期自主智能体 (Agent)、工具调用与自动化工作流打造的大模型，具备 400K 上下文窗口、118 Token/秒生成速度与 66/100 综合智能指数（每百万 Token $1.75 输入 / $14 输出）。
$1.75 / 1M
$14.00 / 1M
400K tokens
78.1%
Devstral 2
自主智能体与工具专家Mistral AI 推出的 Devstral 2 是一款专为长周期自主智能体 (Agent)、工具调用与自动化工作流打造的大模型，具备 262K 上下文窗口、70 Token/秒生成速度与 56/100 综合智能指数（每百万 Token $1.5 输入 / $4.5 输出）。
$1.50 / 1M
$4.50 / 1M
262.1K tokens
58.2%
核心差异一览
- Devstral 2 is 2.3x cheaper for API usage (input + output tokens combined).
- GPT-5.2 Codex has a larger context window (400K vs 262K tokens).
- GPT-5.2 Codex is faster at 118 tok/s vs 70 tok/s.
- GPT-5.2 Codex leads on SWE-bench coding (78.1% vs 58.2%).
| Feature | GPT-5.2 Codex | Devstral 2 |
|---|---|---|
Intelligence Index Composite reasoning & knowledge score (0-100) | 66 | 56 |
Provider | OpenAI | Mistral AI |
Input Price Cost per million input tokens | $1.75 / 1M | $1.50 / 1M |
Output Price Cost per million output tokens | $14.00 / 1M | $4.50 / 1M |
Blended Price Weighted 7:2:1 cache:input:output per 1M | $1.87 / 1M | $0.99 / 1M |
Context Window Maximum input + output tokens | 400K tokens~600 A4 pages of text | 262.1K tokens~393 A4 pages of text |
Max Output Tokens Per single request | 128K | 66K |
Throughput Speed Tokens generated per second | 118 tok/s | 70 tok/s |
SWE-bench Verified Real-world software engineering | 78.1% | 58.2% |
GPQA Diamond Graduate-level reasoning | 78.2% | 66.5% |
Cached Input Price Prompt caching discount per 1M tokens | $0.17 / 1M | $0.35 / 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.
Extremely difficult, closed-ended questions spanning the frontier of human expert knowledge.
Autonomous multi-step tasks executed inside a real command-line terminal environment.
Research-grade coding problems drawn from real scientific and numerical workloads.
Contamination-free competitive programming problems collected continuously over time.
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.
GPT-5.2 Codex · Strengths & Key Advantages
- 提供 400K 上下文窗口与最高 128K 输出容量，核心基准表现优异（SWE-bench Verified 78.1% · GPQA Diamond 78.2%）
- 高达 118 Token/秒的生成吞吐量，首字延迟 (TTFT) 仅约 1280 毫秒
- 极具竞争力的 API 定价（每百万 Token $1.75 输入 / $14 输出），原生支持提示词缓存、函数调用与结构化 JSON 输出
Considerations & Trade-offs
- 闭源专有云端模型，不支持下载权重进行离线本地物理隔离部署
- 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
Devstral 2 · Strengths & Key Advantages
- 提供 262K 上下文窗口与最高 66K 输出容量，核心基准表现优异（SWE-bench Verified 58.2% · GPQA Diamond 66.5%）
- 高达 70 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 | GPT-5.2 Codex | Devstral 2 |
|---|---|---|
Cost per Task Estimated USD to finish one task | $2.20 | $0.65 |
Output Tokens / Task Tokens generated per task | 60K | 42K |
Reasoning Tokens / Task Thinking tokens burned per task | 34K | 20K |
技术规格参数
Architecture, release timeline, capabilities, and modalities.
| Feature | GPT-5.2 Codex | Devstral 2 |
|---|---|---|
| Official API Model ID (SDK) | gpt-5-2-codex | devstral-2-latest |
| API Protocol | OpenAI Responses API & Chat Completions v1 | OpenAI-Compatible REST / vLLM / SGLang / HuggingFace TGI |
| Release Date | 2025-12 | 2025-12 |
| Knowledge Cutoff | August 2025 | October 2025 |
| Architecture | Code-specialised long-horizon reasoning model | Advanced Agentic Code Transformer |
| Parameters | Confidential | 123B |
| Enterprise Rate Limits (RPM / TPM) | 10,000 – 30,000 RPM (Tier 5 Enterprise) · 800,000 – 2,000,000 TPM | 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 ($1.75/1M in) across full 400K window | Flat rate ($1.5/1M in) across full 262K window |
| Prompt Cache Policy & TTL | Automatic prefix caching at $0.175/1M (no extra cache-write fee) | Cached input at $0.35/1M tokens |
| Computer Use / GUI Agent | Supported (Screen coordinate grounding, browser & desktop GUI action loops) | 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 | Native Web Search tool + Hosted Code Interpreter & File Search RAG | Supported via external Tavily/Exa Search & E2B/Docker Sandbox tools |
| Official Fine-Tuning Support | Supervised Fine-Tuning (SFT), Direct Preference (DPO) & Reinforcement Fine-Tuning (RFT) | Full Weights SFT, LoRA, QLoRA, DPO & GRPO RL (Unsloth / Axolotl / Llama-Factory) |
| Multimodal Token Billing | Text BPE (~4 chars/token) · Vision (~258–765 tokens per image tile) | Text BPE (~4 chars/token) |
| Quantization & Min GPU VRAM | Cloud-Hosted Proprietary (FP8 / BF16 Tensor Core Cluster) · Zero Local VRAM Required (100% Serverless Cloud API) | BF16 / FP16, FP8 Dynamic, INT4 AWQ / GPTQ, GGUF (Q4_K_M / Q8_0) · 76 GB VRAM (4-bit Quantized) · 264 GB VRAM (Native BF16) |
| Safety & Frontier Classification | OpenAI Preparedness Framework Audited & System Card Published | Open-Weights System Card + Customizable Guardrail Layer (Llama Guard / ShieldGemma) |
| Enterprise Compliance (ZDR/HIPAA/SOC2) | SOC2 Type II · GDPR · Zero Data Retention (ZDR) & HIPAA BAA Available on Enterprise Tier | 100% Air-Gapped On-Premises / Private VPC · Full HIPAA, GDPR & SOC2 Control |
| Input Modalities | text, vision | text |
| Output Modalities | text | text |
GPT-5.2 Codex vs Devstral 2: 常见问题解答 (FAQ)
Which is more intelligent, GPT-5.2 Codex or Devstral 2?
GPT-5.2 Codex is more intelligent. It scores 66 compared with 56 on the composite Intelligence Index.
Which model is faster?
GPT-5.2 Codex is faster, generating 118 tokens per second compared with 70 tokens per second.
Which model is cheaper to run?
Devstral 2 is cheaper at $0.99 per 1M tokens blended, compared with $1.87 (7:2:1 cache:input:output ratio).
Which model has lower latency?
Devstral 2 has lower latency, with a time to first token of 0.52s compared with 1.28s.
Which model has a larger context window?
GPT-5.2 Codex has a larger context window, supporting 400K tokens compared with 262.1K tokens.
综合结论
综合来看，GPT-5.2 Codex（适合：专为长周期自主智能体 (Agent)、工具调用与自动化工作流打造的任务、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 Devstral 2（适合：专为长周期自主智能体 (Agent)、工具调用与自动化工作流打造的任务、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：GPT-5.2 Codex 在综合智能与编程基准测试中表现更强，GPT-5.2 Codex 拥有更快的 Token 生成速度，而 Devstral 2 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，GPT-5.2 Codex 和 Devstral 2 哪个更强？+
在综合智能指数与编程基准评测中，GPT-5.2 Codex 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。
GPT-5.2 Codex 与 Devstral 2 的 API 价格哪个更便宜？+
GPT-5.2 Codex 的 API 价格为输入 $1.75 / 输出 $14.00（每百万 Token），Devstral 2 为输入 $1.50 / 输出 $4.50。综合混合成本来看，Devstral 2 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？+
GPT-5.2 Codex 支持 400K Token 上下文窗口，而 Devstral 2 支持 262.1K Token。在流式生成速度方面，GPT-5.2 Codex 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
- Devstral 2 vs Claude Sonnet 5.5：基准测试、API 价格与速度全面对比 (2026)对比
- Devstral 2 vs Qwen3-Coder 480B-A35B Instruct：基准测试、API 价格与速度全面对比 (2026)对比
- GPT-5.1 Codex Max vs Devstral 2：基准测试、API 价格与速度全面对比 (2026)对比
- GPT-5.1 Codex vs Devstral 2：基准测试、API 价格与速度全面对比 (2026)对比
- GPT-5.2 Codex vs GPT-5.1 Codex Max：基准测试、API 价格与速度全面对比 (2026)对比
- Aion-3.0 vs Aion-Mini：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: OpenAI 与 Mistral AI 官方 API 文档、模型发布说明及独立基准测试报告 (2026)。