{"zh": "Google Gemini omni 1.1 flash-ext 文生视频模型(上游别名 Omni-Flash-Ext)。走统一的视频生成接口,接受 prompt、时长、分辨率与画面比例,也可传参考图做图生视频。Flash 档主打生成速度。", "en": "Google's Gemini omni 1.1 flash-ext text-to-video model (exposed upstream as Omni-Flash-Ext). It uses the unified video generation endpoint and takes a prompt, duration, resolution and aspect ratio, and can also accept reference images for image-to-video. The Flash tier is built for generation speed."}
{"zh": "Google Gemini omni 1.1 flash-ext 文生视频模型(上游别名 Omni-Flash-Ext)。走统一的视频生成接口,接受 prompt、时长、分辨率与画面比例,也可传参考图做图生视频。Flash 档主打生成速度。", "en": "Google's Gemini omni 1.1 flash-ext text-to-video model (exposed upstream as Omni-Flash-Ext). It uses the unified video generation endpoint and takes a prompt, duration, resolution and aspect ratio, and can also accept reference images for image-to-video. The Flash tier is built for generation speed."}
{"zh": "Vidu 文生音效模型:输入文字描述,生成 2~10 秒的音效,可以精确控制时长和每段声音出现的时间(如「[1s-4s] 屋顶上的雨声」),也能叠加多个音效,适合给视频、游戏与短片配环境音和音效。", "en": "Vidu's text-to-audio sound-effect model: describe a sound in text and get 2-10 seconds of audio, with precise control over duration and when each sound occurs (e.g. '[1s-4s] raindrops on a rooftop') and support for layering multiple effects, for ambience and sound design in videos, games and shorts."}
{"zh": "OpenAI 的传统补全(completion)基座模型,不是对话模型,主要用于微调与遗留脚本。能力远低于现役模型,仅为兼容保留。", "en": "A legacy OpenAI base completion model - not a chat model - mainly used for fine-tuning and legacy scripts. Far weaker than current models and kept only for compatibility."}
{"zh": "Anthropic 位于 Opus 之上的前沿模型 Claude Fable 5（2026-06 发布，后继为 Fable 5.1），首个公开的「Mythos 级」前沿模型，发布时在软件工程、知识工作、视觉与科学研究等几乎所有能力基准上达到最先进水平。100 万上下文，思考始终开启，面向最苛刻的推理与长程智能体任务。", "en": "Anthropic Claude Fable 5 (Jun 2026; succeeded by Fable 5.1), the frontier tier above Opus — the first publicly available 'Mythos-class' frontier model, state-of-the-art at launch across nearly all benchmarks (software engineering, knowledge work, vision, scientific research). 1M context, thinking always on, for the most demanding reasoning and long-horizon agentic work."}
{"zh": "Anthropic 位于 Opus 之上的前沿模型 Claude Fable 5（2026-06 发布，后继为 Fable 5.1），首个公开的「Mythos 级」前沿模型，发布时在软件工程、知识工作、视觉与科学研究等几乎所有能力基准上达到最先进水平。100 万上下文，思考始终开启，面向最苛刻的推理与长程智能体任务。", "en": "Anthropic Claude Fable 5 (Jun 2026; succeeded by Fable 5.1), the frontier tier above Opus — the first publicly available 'Mythos-class' frontier model, state-of-the-art at launch across nearly all benchmarks (software engineering, knowledge work, vision, scientific research). 1M context, thinking always on, for the most demanding reasoning and long-horizon agentic work."}
{"zh": "Anthropic 位于 Opus 之上的前沿模型 Claude Fable 5（2026-06 发布，后继为 Fable 5.1），首个公开的「Mythos 级」前沿模型，发布时在软件工程、知识工作、视觉与科学研究等几乎所有能力基准上达到最先进水平。100 万上下文，思考始终开启，面向最苛刻的推理与长程智能体任务。", "en": "Anthropic Claude Fable 5 (Jun 2026; succeeded by Fable 5.1), the frontier tier above Opus — the first publicly available 'Mythos-class' frontier model, state-of-the-art at launch across nearly all benchmarks (software engineering, knowledge work, vision, scientific research). 1M context, thinking always on, for the most demanding reasoning and long-horizon agentic work."}
{"zh": "Anthropic 位于 Opus 之上的前沿模型 Claude Fable 5（2026-06 发布，后继为 Fable 5.1），首个公开的「Mythos 级」前沿模型，发布时在软件工程、知识工作、视觉与科学研究等几乎所有能力基准上达到最先进水平。100 万上下文，思考始终开启，面向最苛刻的推理与长程智能体任务。", "en": "Anthropic Claude Fable 5 (Jun 2026; succeeded by Fable 5.1), the frontier tier above Opus — the first publicly available 'Mythos-class' frontier model, state-of-the-art at launch across nearly all benchmarks (software engineering, knowledge work, vision, scientific research). 1M context, thinking always on, for the most demanding reasoning and long-horizon agentic work."}
{"zh": "Anthropic Claude Fable 5.1(2026-09-01 发布),位于 Opus 之上的前沿档,面向最苛刻的推理与长周期智能体任务。相比 Fable 5,长时间智能体编程、多步研究,以及文档、表格、幻灯片类工作更强;输入输出价格不变,缓存读取降到原来的四分之一(仅为输入价的 2.5%)。100 万 token 上下文、单次最多输出 12.8 万 token,知识截止 2026-06;自适应思考始终开启,默认力度 high。不支持强制指定 tool_choice。", "en": "Anthropic Claude Fable 5.1 (released September 1, 2026), the frontier tier above Opus for the most demanding reasoning and long-horizon agentic work. Compared with Fable 5 it is stronger at long-running agentic coding, multistep research and document, spreadsheet and slide work, at the same input and output prices, with cache reads cut to a quarter (2.5% of the input price). 1M-token context with up to 128K output per response and a June 2026 knowledge cutoff; adaptive thinking is always on with high default effort. Forced tool_choice is not supported."}
{"zh": "Anthropic Claude Fable 5.1(2026-09-01 发布),位于 Opus 之上的前沿档,面向最苛刻的推理与长周期智能体任务。相比 Fable 5,长时间智能体编程、多步研究,以及文档、表格、幻灯片类工作更强;输入输出价格不变,缓存读取降到原来的四分之一(仅为输入价的 2.5%)。100 万 token 上下文、单次最多输出 12.8 万 token,知识截止 2026-06;自适应思考始终开启,默认力度 high。不支持强制指定 tool_choice。", "en": "Anthropic Claude Fable 5.1 (released September 1, 2026), the frontier tier above Opus for the most demanding reasoning and long-horizon agentic work. Compared with Fable 5 it is stronger at long-running agentic coding, multistep research and document, spreadsheet and slide work, at the same input and output prices, with cache reads cut to a quarter (2.5% of the input price). 1M-token context with up to 128K output per response and a June 2026 knowledge cutoff; adaptive thinking is always on with high default effort. Forced tool_choice is not supported."}
{"zh": "Anthropic Claude Fable 5.1(2026-09-01 发布),位于 Opus 之上的前沿档,面向最苛刻的推理与长周期智能体任务。相比 Fable 5,长时间智能体编程、多步研究,以及文档、表格、幻灯片类工作更强;输入输出价格不变,缓存读取降到原来的四分之一(仅为输入价的 2.5%)。100 万 token 上下文、单次最多输出 12.8 万 token,知识截止 2026-06;自适应思考始终开启,默认力度 high。不支持强制指定 tool_choice。", "en": "Anthropic Claude Fable 5.1 (released September 1, 2026), the frontier tier above Opus for the most demanding reasoning and long-horizon agentic work. Compared with Fable 5 it is stronger at long-running agentic coding, multistep research and document, spreadsheet and slide work, at the same input and output prices, with cache reads cut to a quarter (2.5% of the input price). 1M-token context with up to 128K output per response and a June 2026 knowledge cutoff; adaptive thinking is always on with high default effort. Forced tool_choice is not supported."}
{"zh": "Anthropic 最快、最便宜的模型 Claude Haiku 4.5（2025-10 发布），20 万上下文、64K 输出，以约 1/3 成本、两倍以上速度达到接近 Sonnet-4 级编程水平，适合延迟与成本敏感的对话、分类与廉价子任务。", "en": "Anthropic's fastest, cheapest model, Claude Haiku 4.5 (Oct 2025), 200K context and 64K output — near Sonnet-4-level coding at ~1/3 the cost and 2x+ the speed, for latency- and cost-sensitive chat, classification and cheap agent sub-tasks."}