OpenAI
GPT-6.2 Astra
最大上下文
1.1M
最大输出
128K
发布日期
暂无
gpt-5.4-miniGPT-5.4 Mini 是 GPT-5.4 家族中更小、更快的版本。OpenAI 文档将 mini 模型定位为低延迟工作负载,同时保留代码、工具调用、多模态推理和较强指令遵循。它适合边界清晰的生产任务、子 Agent,以及需要大量快速调用的应用。
上下文窗口
400K Token
最大输出
128K Token
发布日期
2026年3月17日
模态
这是价格描述
推理
工具调用
Temperature 参数
附件
知识库
Endpoint 协议
GPT-5.4 Mini 以更快、更适合高吞吐任务的形态提供可调推理、编程和视觉计算机操作能力。
在速度和效率更高的模型中保留 GPT-5.4 级推理能力,适合处理大量生产请求。
能够完成代码生成、分析和多文件修改,同时满足低延迟开发工作流的需求。
理解截图和界面状态,并据此判断浏览器及桌面环境中的操作步骤。
GPT-5.4 Mini 适合高容量编程辅助、视觉计算机自动化和需要快速执行的 Agent 委派任务。
在开发工具中处理大量代码生成、解释和审查请求,同时保持实用的响应速度。
解读界面截图、跟踪应用状态,并完成可重复的多步骤浏览器或桌面任务。
在大型工作流中执行边界明确的研究、编程或分析任务,并向协调 Agent 返回结构化结果。
import OpenAI from "openai"
const client = new OpenAI({
apiKey: process.env.TOKENHUB_API_KEY,
baseURL: "https://us-api.tokenhub.com/v1",
})
const result = await client.chat.completions.create({})
console.log(result.choices[0]?.message?.content)| 指数分数 | ||
|---|---|---|
| Artificial Analysis 综合能力指数 | Artificial Analysis 综合能力评分 | 40 |
| Artificial Analysis 代码能力指数 | Artificial Analysis 软件任务评分 | 51.5 |
| 知识与推理 | ||
| GPQA | 高阶科学问题求解 | 87.5% |
| HLE | 广泛专家级考试集合 | 26.6% |
| 代码与工程 | ||
| SciCode | 科学计算代码挑战 | 49.9% |
| Terminal-Bench Hard | 高难度终端任务执行 | 52.3% |
| 指令遵循与 Agent 任务 | ||
| IFBench | 指令约束遵循 | 73.3% |
| AA-LCR | 长上下文推理 | 69.3% |
| τ²-Bench | Agent 工作流任务 | 83.3% |
指标来源 Artificial Analysis
GPT-5.4 Mini 的能力、适用场景、限制和 TokenHub 使用指南。
GPT-5.4 Mini 是 OpenAI 面向快速高效的编程、工具调用和多模态任务的模型。
适合日常编程辅助、大量工具调用的自动化、图像和视频理解,尤其适合重视响应速度和成本效率的场景。
核心优势是以更低延迟和成本提供强大的编程与工具能力。
它用部分峰值质量换取更好的速度或成本。 更看重最高回答质量时,可考虑 GPT-5.4。
使用 TokenHub 显示的精确标识符,并按账户文档确认当前功能。
通过一个 API Key 接入 GPT-5.4 Mini 与更多 AI 模型。
媒体与讨论
精选与此模型相关的公开视频和帖子。
X (Twitter)
OpenAI ·
GPT-5.4 mini is available today in ChatGPT, Codex, and the API. Optimized for coding, computer use, multimodal understanding, and subagents. And it’s 2x faster than GPT-5 mini. https://openai.com/index/introducing-gpt-5-4-mini-and-nano/ https://t.co/sirArgn37L
https://pbs.twimg.com/media/HDoN4PhasAAinj_.png查看原文OpenAI ·
GPT-5.4 Thinking and GPT-5.4 Pro are rolling out now in ChatGPT. GPT-5.4 is also now available in the API and Codex. GPT-5.4 brings our advances in reasoning, coding, and agentic workflows into one frontier model. https://t.co/1hy6xXLAmJ
https://pbs.twimg.com/media/HCqogsobgAAAo2U.png查看原文OpenAI ·
GPT-5.4 is our most factual and efficient model: fewer tokens, faster speed. In ChatGPT, GPT-5.4 Thinking has improved deep web research, better context retention when it thinks for longer—and oh—you can now interrupt the model and add instructions or adjust its direction https://t.co/a0ayi6aOYV
https://pbs.twimg.com/media/HCqrhVeaUAYk72Y.jpg查看原文Reddit
The REAL CASE for GPT 5.4 mini - According to Open AI.
kknd1991 ·
查看原文GPT 5.4 default vs 5.4 mini high - performed similarly, but is there a large cost difference?
TruthTellerTom ·
展开全文
I’ve been using GPT-5.4 Mini on default for both planning and building. Whenever something feels more complex or needs broader context, I switch to regular GPT-5.4, also on default. Up to now, I never paid much attention to the thinking setting. I just left it on default and worked that way. This strategy has kept my token use low instead of just using 5.4 all the time. Today I ran into a case where GPT-5.4 Mini on default gave me an answer that was technically workable, but it missed the bigger context of the issue. The solution might have fixed the immediate problem but would have caused issues somewhere else in the code. Thankfully I caught it before moving forward. I then rolled back and reran the same kind of planning/analysis using regular GPT-5.4. That version understood the problem properly and gave me the correct solution. I repeated the comparison a few times, going back and forth between Mini default and regular 5.4, and the pattern stayed the same: regular 5.4 consistently understood the broader context better, while Mini default kept missing it and suggesting weaker solutions. After that, I tried GPT-5.4 Mini on high thinking. That changed things. Mini high was suddenly able to follow the broader context much better and started giving answers much closer to regular 5.4. I repeated that a few times too, and Mini high stayed much more on point. So now I’m thinking of changing my workflow to this: GPT-5.4 Mini default for simpler work GPT-5.4 Mini high for tasks that are more complex or need wider context regular GPT-5.4 only when truly needed My question is: in real-world use, how much savings am I actually getting with Mini high compared to regular 5.4? At that point, is Mini high still meaningfully cheaper, or does it get close enough that I might as well just use regular 5.4? I’d also be curious how other people decide when to step up from Mini default to Mini high vs going straight to full 5.4.
Anyone tried GPT-5.4 Mini? Worth it?
Plus_Leadership_6886 ·
I’ve recently started noticing GPT-5.4 Mini showing up, but I haven’t really seen many people talking about it yet. Has anyone here actually used it? How does it compare to previous versions in terms of quality? Also, does it make any real difference when it comes to usage limits or rate limits?
查看原文YouTube
GPT-5.4 Mini & Nano: OpenAI's FASTEST AND Most Capable Models Yet!
Universe of AI
查看原文GPT 5.4 Mini in 5 mins!
1littlecoder
查看原文OpenAI launches GPT 5.4 Mini & Nano: 92% less cost
Sundas Khalid
查看原文