AI Models

GPT-5.6

GPT-5.6

Closed

OpenAI's latest 2026 general and reasoning flagship.

Model background: GPT-5.6 延续统一模型路线，把通用对话、代码、视觉和 Agent 工具调用收束到同一套产品接口。

GPT-5

GPT-5

Closed

OpenAI's flagship model with adaptive reasoning and fast responses.

Model background: GPT-5 是 OpenAI 面向通用任务的新一代旗舰模型，重点放在复杂推理、代码和多模态协作，并在回答速度与思考深度之间自动切换。

GPT-5.5

GPT-5.5

Closed

A GPT-5 upgrade tuned for developers and knowledge work.

Model background: GPT-5.5 将复杂任务的稳定性、长上下文处理和工具使用体验推向更成熟的生产阶段。

Claude Sonnet 4.6

Claude Sonnet 4.6

Closed

Anthropic's latest balanced model for coding and agents.

Model background: Sonnet 4.6 延续 Sonnet 的速度优势，将更长的任务链、代码修改和工具调用纳入日常开发流程。

Claude Opus 4.5

Claude Opus 4.5

Closed

Anthropic's flagship model for demanding coding and agent tasks.

Model background: Claude Opus 4.5 是 Anthropic 面向高难度知识工作与 Agent 流程的旗舰模型，强调复杂代码任务、长程规划和高质量写作。

GPT-5.4

GPT-5.4

Closed

A GPT-5 release strengthened for vision and document understanding.

Model background: GPT-5.4 的文档和视觉能力让模型更像一名可以检查表格、截图和文件的协作者。

Gemini 2.5 Pro

Gemini 2.5 Pro

Closed

Google's flagship multimodal model with an exceptionally long context and video understanding.

Model background: Gemini 2.5 Pro 是 Google 的高级多模态模型，主打长上下文、复杂推理以及图像和视频理解，并可通过 Gemini API 使用。

DeepSeek-R1

DeepSeek-R1

Open

DeepSeek's open reasoning model for math, coding, and complex problem decomposition.

Model background: DeepSeek-R1 是 DeepSeek 发布的开源推理模型，重点强化数学、代码和复杂问题拆解能力，也适合在本地或第三方平台部署。

GPT-4

GPT-4

ClosedRetired

The GPT-4 milestone that pushed frontier models into products.

Model background: 2023 年发布时，GPT-4 让多模态、代码与专业考试能力第一次被大规模用户感知，成为后来所有模型的标尺。

Claude Sonnet 4

Claude Sonnet 4

Closed

Anthropic's high-performance model for coding, long documents, and complex reasoning.

Model background: Claude Sonnet 4 是 Anthropic Claude 家族中的均衡型号，在响应速度、代码能力、长文档处理和复杂推理之间取得平衡。

GPT-4o

GPT-4o

Closed

OpenAI's multimodal model for general Q&A, vision, and tool use.

Model background: GPT-4o 是 OpenAI 的 omni 多模态模型，围绕文本、图像和语音的实时交互设计，适合通用助手、视觉理解与工具调用。

Claude 3.5 Sonnet

Claude 3.5 Sonnet

ClosedRetired

The Claude classic that brought coding agents into the mainstream.

Model background: 2024 年 Claude 3.5 Sonnet 在代码和视觉任务上连续刷新榜单，许多 AI 编程产品以它作为默认大脑。

GPT-3.5 Turbo

GPT-3.5 Turbo

ClosedRetired

The model that put chat completions in every developer's toolbox.

Model background: ChatGPT 爆发后，GPT-3.5 Turbo 以低价和速度推动了插件、客服和无数个人项目的第一波浪潮。

Sora

Sora

Closed

OpenAI's text-to-video model that pushed generative video into the mainstream.

Model background: Sora 2024 年的长镜头演示重新定义了公众对视频生成连贯性和物理世界模拟的期待。

gpt-oss-120B

gpt-oss-120B

Open

OpenAI's open-weight reasoning model for self-hosting.

Model background: gpt-oss-120B 是 OpenAI 开放模型路线的重要节点，把旗舰级推理能力带进可自托管生态。

DeepSeek-V3

DeepSeek-V3

Open

DeepSeek's open general model for cost-effective chat and coding.

Model background: DeepSeek-V3 是 DeepSeek 的开源通用模型，定位于日常对话、代码和 API 服务，在能力与调用成本之间强调性价比。

OpenAI o1

OpenAI o1

ClosedRetired

A reasoning model that trades latency for stronger math and science reliability.

Model background: o1 把‘先思考再回答’从研究概念带进产品，开启了 2024 年推理模型竞赛。

Gemini 1.5 Pro

Gemini 1.5 Pro

ClosedRetired

The Google model that popularized million-token context windows.

Model background: Gemini 1.5 Pro 的百万上下文演示改变了人们对‘一次能读多少’的想象，推动长上下文成为标配。

Whisper

Whisper

Open

An openly released multilingual speech recognizer that became a de facto standard.

Model background: Whisper 以大规模弱监督训练把多语言转写带到开发者手中，催生了字幕、会议和语音输入生态。

Gemini 3.5 Transcribe

Gemini 3.5 Transcribe

Closed

Google's 2026 model for realtime and recorded transcription.

Model background: Gemini 3.5 Transcribe 把自我纠正、口头填充词、格式化和专业词汇处理带进同一套语音接口。

Llama 3.1

Llama 3.1

Open

One of the most widely deployed general models in the open-weight ecosystem.

Model background: Llama 3.1 将 405B 级别能力以开放权重形式带给社区，催生了本地推理、量化和微调工具链。

Gemini Omni 1.1 Flash

Gemini Omni 1.1 Flash

Closed

A video model with extension, first/last-frame, and 4K workflows.

Model background: Omni 1.1 Flash 把 360p 草稿、首尾帧控制和 4K 放大串成可检查的视频制作流程。

Llama 4 Maverick

Llama 4 Maverick

Open

The Llama 4 open model for high-quality multimodal work.

Model background: Llama 4 Maverick 把 MoE 和原生视觉能力结合，继续扩大开放权重模型在产品中的可用范围。

OpenAI o3

OpenAI o3

Closed

OpenAI's deep-reasoning model for math, science, and complex planning.

Model background: o3 是 OpenAI 的深度推理系列模型，针对数学、科学、代码和多步规划等需要更长思考链路的任务进行优化。