{"auto_groups":["default","Codex Team","Codex Pro","Codex Plus"],"data":[{"model_name":"o3","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"],"pricing_version":"5a90f2b86c08bd983a9a2e6d66c255f4eaef9c4bc934386d2b6ae84ef0ff1f1f"},{"model_name":"claude-opus-4-6","description":"Claude Opus 4.6 是 Anthropic 于 2026 年 2 月推出的顶奢旗舰模型，专为“端到端企业自动化”设计。它不仅是 Opus 4.5 的直接升级版，更在长文本处理上实现了量级突破。Opus 4.6 的核心竞争力在于其近乎人类的“自主权”，它能独立完成从“信息检索”到“深度分析”再到“最终交付物生成（如 PPT 或 Excel）”的全流程。该模型在网络安全领域表现惊人，无需特殊工具即可发现代码中隐藏数十年的 0-day 漏洞。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Vertex","Kiro Business","Claude Code Max","Claude Mix","Claude Code","Claude Aws Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-image-2.5","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.075,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-3-5-sonnet-20241022","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Mix"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"kimi-k2.5","description":"Kimi-k2.5 是月之暗面（Moonshot AI）迄今为止最全能、最智能的旗舰级多模态模型。该模型采用了原生的多模态架构设计，将视觉理解、逻辑推理、代码生成及 Agent 任务处理能力深度集成。相比前代 K2，k2.5 在 Agent 自动化任务上取得了显著突破，支持超过 300 步的复杂工具调用，能自主完成数据抓取、代码执行及深度研究报告撰写。其独特的“思考与非思考”双模式设计，允许模型在面对复杂逻辑时进行深度长链条推理（Long-horizon Reasoning），而在常规对话中保持极快的响应速度。","tags":"对话,识图,多模态,推理","vendor_id":5,"quota_type":0,"model_ratio":2,"model_price":0,"owner_by":"","completion_ratio":5.25,"enable_groups":["Chinese Model","default"],"supported_endpoint_types":["openai"]},{"model_name":"glm-5.3","vendor_id":3,"quota_type":0,"model_ratio":4,"model_price":0,"owner_by":"","completion_ratio":3.5,"cache_ratio":0.25,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"doubao-seedance-1-0-pro-250528","vendor_id":23,"quota_type":1,"model_ratio":0,"model_price":0.1,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_usage_schema":{"resolution":{"enum":["480p","720p","1080p","4k"],"description":{"en":"Output video resolution","zh":"输出视频分辨率"},"enumLabels":{"1080p":{"en":"1080p","zh":"1080p"},"480p":{"en":"480p","zh":"480p"},"4k":{"en":"4k","zh":"4k"},"720p":{"en":"720p","zh":"720p"}}},"tokens":{"type":"number","unit":"token","description":{"en":"Billing token unit price","zh":"计费 Token 单价"}},"video_input":{"enum":["none","video"],"description":{"en":"Reference video input","zh":"参考视频输入"},"enumLabels":{"none":{"en":"No reference video","zh":"无参考视频"},"video":{"en":"With reference video","zh":"有参考视频"}}}},"billing_usage_examples":[{"label":"480p · 5s","facts":{"resolution":"480p","tokens":48038,"video_input":"none"}},{"label":"720p · 5s","facts":{"resolution":"720p","tokens":108000,"video_input":"none"}},{"label":"1080p · 5s","facts":{"resolution":"1080p","tokens":243000,"video_input":"none"}},{"label":"4k · 5s","facts":{"resolution":"4k","tokens":972000,"video_input":"none"}},{"label":"720p · 10s","facts":{"resolution":"720p","tokens":216000,"video_input":"none"}},{"label":"720p · 5s (+4s 输入视频)","facts":{"resolution":"720p","tokens":194400,"video_input":"video"}}]},{"model_name":"gemini-3.1-flash-image-preview","vendor_id":6,"quota_type":1,"model_ratio":0,"model_price":0.15,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"claude-opus-5-5","vendor_id":1,"quota_type":0,"model_ratio":2,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.05,"create_cache_ratio":2,"enable_groups":["Claude Code Max","default","Claude Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-opus-4-1-20250805-thinking","description":"Claude Opus 4.1 是 Anthropic 对其顶级旗舰模型 Opus 4 的精准性能优化版。它被设计为 Opus 4 的平替版，在保持相同 API 架构和价格的同时，显著提升了在真实世界编程（Real-world coding）和长程代理任务中的表现。该模型在 SWE-bench Verified 测试中取得了 74.5% 的高分，超越了初代 Opus 4。它的核心突破在于增强了对多文件重构的精准度和错误追踪能力，并引入了“混合推理”机制，支持开发者在快速响应和深度步进思考（Extended Thinking）之间灵活切换，是企业级自动化代理和科研辅助的首选。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":7.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Mix","Claude Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-sonnet-4-20250514-thinking","description":"Claude Sonnet 4 是 Anthropic 4.0 系列的核心型号，旨在接替广受好评的 Claude 3.5 Sonnet。它引入了革命性的“混合推理”架构，允许模型在“即时响应”和“深度思考（Extended Thinking）”之间根据任务复杂度动态切换。Sonnet 4 在编程能力、指令遵循和长文本记忆力方面实现了跨越式提升，其在 SWE-bench 编程基准测试中达到了 72.7% 的历史高位。它不仅能处理极其复杂的代码重构，还能在长达 100 万 token 的超大上下文中实现近乎完美的记忆调取，是目前企业级 AI 代理和专业开发者的主流生产力工具。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"deepseek-ocr","vendor_id":4,"quota_type":0,"model_ratio":0.2,"model_price":0,"owner_by":"","completion_ratio":4.5,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3-flash-preview","description":"Google Gemini 3 系列的高速、高性价比推理模型。它采用了全新的“思维级别”可调架构，在保持 Flash 系列毫秒级响应速度的同时，通过原生思维链（CoT）支持，在编程能力和多步逻辑拆解上达到了前代 Pro 模型的水平。","tags":"对话,多模态,推理,工具","vendor_id":6,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.25,"enable_groups":["Gemini Premium","Gemini AI Studio","default","Gemini Cli","Gemini Antigravity","Gemini Vertex"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"grok-420-fast-deepsearch-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.0075,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-2.5-flash-preview-tts","vendor_id":6,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":20,"cache_ratio":0.1,"enable_groups":["Gemini Vertex","Gemini Premium"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"deepseek-v4.1-flash","vendor_id":4,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.1,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"minimax-m2.7","vendor_id":20,"quota_type":0,"model_ratio":1.05,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"glm-4.6","vendor_id":3,"quota_type":0,"model_ratio":1.575,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["Chinese Model","default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-sonnet-4-6","description":"Sonnet 4.6 是 Anthropic 迄今为止功能最强大的 Sonnet 系列产品，在编码、代理和专业工作方面均展现出卓越的性能。它尤其擅长迭代开发、复杂代码库导航、端到端项目管理（含内存管理）、文档创建以及在 Web 质量保证和工作流自动化方面的出色计算机应用。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Claude Aws Code","default","Claude Vertex","Kiro Business","Claude Code Max","Claude Mix","Claude Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gemini-2.5-flash-lite","description":"Gemini 2.5 系列中速度最快、性价比最高的多模态推理模型。它针对高吞吐量、低延迟任务进行了优化，例如翻译、分类和实时路由。作为一款“思维模型”，它支持可选的推理预算，能够以远低于以往 Flash 芯片的成本提供更优的逻辑性能。","tags":"对话,多模态,推理,工具","vendor_id":6,"quota_type":0,"model_ratio":0.05,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["Gemini Cli","Gemini Vertex","Gemini Premium","Gemini AI Studio","default","Gemini Antigravity"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gemini-3.1-flash-tts-preview","vendor_id":6,"quota_type":0,"model_ratio":0.5,"model_price":0,"owner_by":"","completion_ratio":4,"audio_ratio":1,"audio_completion_ratio":20,"enable_groups":["Gemini Premium","Gemini Vertex"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"glm-5","description":"GLM-5 是智谱新一代的旗舰基座模型，面向 Agentic Engineering 打造，能够在复杂系统工程与长程 Agent 任务中提供可靠生产力。在 Coding 与 Agent 能力上，GLM-5 取得开源 SOTA 表现，在真实编程场景的使用体感逼近 Claude Opus 4.5，擅长复杂系统工程与长程 Agent 任务，是通用 Agent 助手的理想基座。","tags":"对话","vendor_id":3,"quota_type":0,"model_ratio":3,"model_price":0,"owner_by":"","completion_ratio":3.666666666667,"enable_groups":["Chinese Model","default"],"supported_endpoint_types":["openai"]},{"model_name":"glm-5.1","vendor_id":3,"quota_type":0,"model_ratio":3,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"doubao-seedream-5-0","vendor_id":23,"quota_type":1,"model_ratio":0,"model_price":0.05,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"grok-420-fast-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.0075,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-fable-5-1","vendor_id":1,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Claude Code Max","default","Claude Aws Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-image-2-pro","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.1-flash-lite","vendor_id":6,"quota_type":0,"model_ratio":0.125,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Gemini Cli","Gemini Vertex","Gemini Premium","default","Gemini Antigravity"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"grok-420-thinking-deepsearch-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.0075,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"qwen3-coder-480b","vendor_id":21,"quota_type":0,"model_ratio":3.6,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"glm-4.7","description":"GLM-4.7 是智谱新一代旗舰模型，总参数量 355B，激活参数量 32B，在通用对话、推理和智能体能力方面全面升级。GLM-4.7 增强了 Interleaved Thinking（交错思考），并引入了 Preserved Thinking（保留思考）和 Turn-level Thinking（轮级思考），使复杂任务更加稳定可控。在 agentic coding 和终端任务方面，GLM-4.7 在 SWE-bench 达到 73.8%，在 SWE-bench Multilingual 达到 66.7%，在 Terminal Bench 2.0 达到 41%。GLM-4.7 在\"Vibe Coding\"上也有显著提升，能生成更简洁现代的网页，以及布局和尺寸更精准的幻灯片","tags":"对话","vendor_id":3,"quota_type":0,"model_ratio":2,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"claude-3-7-sonnet-20250219","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Mix"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-sonnet-4-5-20250929-thinking","description":"Claude Sonnet 4.5 是 Anthropic 开发的尖端智能模型，专门为现实世界中的复杂代理（Agents）和编程工作流而优化。它是目前编程能力最强的模型，在 SWE-bench Verified 等基准测试中刷新了纪录。该模型在计算机操作（Computer Use）、网络搜索、金融分析和网络安全方面展现了跨越式的提升。其核心突破在于能够维持长达 30 小时以上的自主任务专注度，并支持全新的“思维链（Extended Thinking）”模式，允许模型在输出前进行深度思考。它在保持高性能的同时，沿用了 3.5/4.0 时代的定价体系，是目前最具性价比的生产力模型。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Claude Code","default"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-5.4-openai-compact","vendor_id":2,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Codex Pro"],"supported_endpoint_types":["openai"]},{"model_name":"claude-opus-4-5-20251101-thinking","description":"Claude 4.5 Opus 是 Anthropic 目前功能最强大的模型。它在处理极其复杂的任务（如尖端科学研究、复杂的编码架构设计和深层次的战略分析）方面表现卓越。相比之前的版本，它显著提升了逻辑推理的连贯性，并大幅减少了幻觉。Opus 版本特别强化了对多模态输入的跨表分析能力，能够同时理解并关联数千页文档、代码库以及高清视频流中的细节。它还引入了更加细粒度的“宪法 AI”约束，使其在保持强大能力的同时，更加安全且符合人类价值观。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Vertex","Claude Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gemini-3.8-flash","vendor_id":6,"quota_type":0,"model_ratio":0.375,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"enable_groups":["default","Gemini Antigravity","Gemini Cli","Gemini Vertex","Gemini Premium"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"grok-3-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.005,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-sonnet-4-5-20250929","description":"Claude Sonnet 4.5 是 Anthropic 开发的尖端智能模型，专门为现实世界中的复杂代理（Agents）和编程工作流而优化。它是目前编程能力最强的模型，在 SWE-bench Verified 等基准测试中刷新了纪录。该模型在计算机操作（Computer Use）、网络搜索、金融分析和网络安全方面展现了跨越式的提升。其核心突破在于能够维持长达 30 小时以上的自主任务专注度，并支持全新的“思维链（Extended Thinking）”模式，允许模型在输出前进行深度思考。它在保持高性能的同时，沿用了 3.5/4.0 时代的定价体系，是目前最具性价比的生产力模型。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Claude Aws Code","default","Kiro Business","Claude Code Max","Claude Mix","Claude Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-opus-4-1-20250805","description":"Claude Opus 4.1 是 Anthropic 对其顶级旗舰模型 Opus 4 的精准性能优化版。它被设计为 Opus 4 的平替版，在保持相同 API 架构和价格的同时，显著提升了在真实世界编程（Real-world coding）和长程代理任务中的表现。该模型在 SWE-bench Verified 测试中取得了 74.5% 的高分，超越了初代 Opus 4。它的核心突破在于增强了对多文件重构的精准度和错误追踪能力，并引入了“混合推理”机制，支持开发者在快速响应和深度步进思考（Extended Thinking）之间灵活切换，是企业级自动化代理和科研辅助的首选。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":7.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Mix","Claude Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-4-turbo-2024-04-09","vendor_id":2,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":3,"enable_groups":["Azure All","default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.5-flash-lite","vendor_id":6,"quota_type":0,"model_ratio":0.15,"model_price":0,"owner_by":"","completion_ratio":8.333333333333,"cache_ratio":0.1,"enable_groups":["Gemini Antigravity","Gemini Cli","Gemini Vertex","Gemini Premium","default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"deepseek-r1","description":"deepseek-r1 是 DeepSeek 专为解决复杂逻辑、数学、编程及长程规划任务而设计的推理模型 API 别名。与追求对话流畅度的 deepseek-chat 不同，该模型在输出最终答案前，会强制生成一段名为 Chain of Thought (CoT，思维链) 的推理过程。在 2026 年初，该接口主要指向经过大规模强化学习（RL）迭代后的 DeepSeek-R1 旗舰版本。它不仅在 AIME 和 Codeforces 等硬核测试中对标 OpenAI o1/o3，还引入了“自我反思”机制，能够识别并修正推理过程中的错误，是构建自动化研发、科学计算及高复杂性 Agent 的核心大脑。","tags":"对话,多模态,工具","vendor_id":4,"quota_type":0,"model_ratio":2,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"qwen3-reranker-8b","vendor_id":21,"quota_type":0,"model_ratio":0.125,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4o","vendor_id":2,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.5,"enable_groups":["default","Azure All","Azure High"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4o-2024-11-20","vendor_id":2,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.5,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.1","description":"OpenAI 于 2025 年 11 月发布的 GPT-5 系列升级版。它将模型拆分为 Instant（即时） 与 Thinking（思考） 两个变体，并通过自动路由系统实现无缝切换。GPT-5.1 相比初代 GPT-5 拥有更温暖、更具“人性化”的对话语气，并显著提升了指令遵循能力和长链路任务的逻辑一致性。","tags":"对话,识图,多模态","vendor_id":2,"quota_type":0,"model_ratio":0.625,"model_price":0,"owner_by":"","completion_ratio":8,"enable_groups":["Azure All","Azure High","default"],"supported_endpoint_types":["openai","openai responses"]},{"model_name":"grok-4.3-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.02,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-sonnet-5","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Kiro Business","Claude Code Max","Claude Mix","Claude Code","Claude Aws Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-5.4-mini","vendor_id":2,"quota_type":0,"model_ratio":0.375,"model_price":0,"owner_by":"","completion_ratio":6,"enable_groups":["Azure All","Codex Pro","Azure High","default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.6-terra","vendor_id":2,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["Codex Team","Codex Plus","Azure High","default","Codex Pro"],"supported_endpoint_types":["openai"]},{"model_name":"claude-opus-4-7-thinking","vendor_id":1,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"glm-5.3-flash","vendor_id":3,"quota_type":0,"model_ratio":0.4,"model_price":0,"owner_by":"","completion_ratio":3.5,"cache_ratio":0.2875,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5-mini","vendor_id":2,"quota_type":0,"model_ratio":0.125,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["Azure All","Azure High","default"],"supported_endpoint_types":["openai"]},{"model_name":"o4-mini","quota_type":0,"model_ratio":0.55,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["Azure All","default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-opus-4-7","vendor_id":1,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Claude Code","Claude Aws Code","default","Claude Vertex","Kiro Business","Claude Code Max","Claude Mix"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"grok-4.5-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.02,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-3.5-turbo-0125","vendor_id":2,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4.1-mini","vendor_id":2,"quota_type":0,"model_ratio":0.2,"model_price":0,"owner_by":"","completion_ratio":2,"cache_ratio":0.25,"enable_groups":["default","Azure All","Azure High"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.6-flash","vendor_id":6,"quota_type":0,"model_ratio":0.75,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"enable_groups":["Gemini Vertex","Gemini Premium","Gemini AI Studio","default","Gemini Antigravity","Gemini Cli"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"kimi-k3","vendor_id":5,"quota_type":0,"model_ratio":10,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4.1-nano","vendor_id":2,"quota_type":0,"model_ratio":0.05,"model_price":0,"owner_by":"","completion_ratio":2,"cache_ratio":0.25,"enable_groups":["default","Azure All","Azure High"],"supported_endpoint_types":["openai"]},{"model_name":"doubao-seedance-1-5-pro-251215","vendor_id":23,"quota_type":1,"model_ratio":0,"model_price":0.1,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_usage_schema":{"resolution":{"enum":["480p","720p","1080p","4k"],"description":{"en":"Output video resolution","zh":"输出视频分辨率"},"enumLabels":{"1080p":{"en":"1080p","zh":"1080p"},"480p":{"en":"480p","zh":"480p"},"4k":{"en":"4k","zh":"4k"},"720p":{"en":"720p","zh":"720p"}}},"tokens":{"type":"number","unit":"token","description":{"en":"Billing token unit price","zh":"计费 Token 单价"}},"video_input":{"enum":["none","video"],"description":{"en":"Reference video input","zh":"参考视频输入"},"enumLabels":{"none":{"en":"No reference video","zh":"无参考视频"},"video":{"en":"With reference video","zh":"有参考视频"}}}},"billing_usage_examples":[{"label":"480p · 5s","facts":{"resolution":"480p","tokens":48038,"video_input":"none"}},{"label":"720p · 5s","facts":{"resolution":"720p","tokens":108000,"video_input":"none"}},{"label":"1080p · 5s","facts":{"resolution":"1080p","tokens":243000,"video_input":"none"}},{"label":"4k · 5s","facts":{"resolution":"4k","tokens":972000,"video_input":"none"}},{"label":"720p · 10s","facts":{"resolution":"720p","tokens":216000,"video_input":"none"}},{"label":"720p · 5s (+4s 输入视频)","facts":{"resolution":"720p","tokens":194400,"video_input":"video"}}]},{"model_name":"claude-3-5-sonnet-20240620","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Mix"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-4-turbo","vendor_id":2,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":3,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"o3-mini","vendor_id":2,"quota_type":0,"model_ratio":0.55,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.5,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.7-flash","vendor_id":6,"quota_type":0,"model_ratio":0.375,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"enable_groups":["Gemini AI Studio","default","Gemini Antigravity","Gemini Cli","Gemini Vertex","Gemini Premium"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"text-embedding-3-small","quota_type":0,"model_ratio":0.01,"model_price":0,"owner_by":"","completion_ratio":0,"enable_groups":["Azure High"],"supported_endpoint_types":["openai"]},{"model_name":"minimax-m2.1","description":"MiniMax-M2.1 是在 M2 架构基础上的全方位进化版本。它最大的突破是实现了 “多语言编程 (Multilingual Programming)” 的 SOTA 性能，不仅精通 Python，还在 Rust、Go、Swift、Kotlin、C++ 等多门语言上达到了行业领先水平。该模型继续采用 230B 总参数/10B 激活参数的 MoE 架构，但引入了更紧凑的 “简洁思维链 (Concise Thinking)” 机制，使其回复速度更快、Token 消耗降低了约 30%。它被设计为能直接驱动复杂智能体（Agent）的高级大脑，能够完成从后端逻辑设计到原生 iOS/Android App 交付的全链路任务。","icon":"MiniMax","tags":"对话,推理,工具","vendor_id":20,"quota_type":0,"model_ratio":1.05,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"doubao-seedance-2-0-260128","vendor_id":23,"quota_type":0,"model_ratio":11.5,"model_price":0,"owner_by":"","completion_ratio":0.04347826087,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_usage_schema":{"resolution":{"enum":["480p","720p","1080p","4k"],"description":{"en":"Output video resolution","zh":"输出视频分辨率"},"enumLabels":{"1080p":{"en":"1080p","zh":"1080p"},"480p":{"en":"480p","zh":"480p"},"4k":{"en":"4k","zh":"4k"},"720p":{"en":"720p","zh":"720p"}}},"tokens":{"type":"number","unit":"token","description":{"en":"Billing token unit price","zh":"计费 Token 单价"}},"video_input":{"enum":["none","video"],"description":{"en":"Reference video input","zh":"参考视频输入"},"enumLabels":{"none":{"en":"No reference video","zh":"无参考视频"},"video":{"en":"With reference video","zh":"有参考视频"}}}},"billing_usage_examples":[{"label":"480p · 5s","facts":{"resolution":"480p","tokens":48038,"video_input":"none"}},{"label":"720p · 5s","facts":{"resolution":"720p","tokens":108000,"video_input":"none"}},{"label":"1080p · 5s","facts":{"resolution":"1080p","tokens":243000,"video_input":"none"}},{"label":"4k · 5s","facts":{"resolution":"4k","tokens":972000,"video_input":"none"}},{"label":"720p · 10s","facts":{"resolution":"720p","tokens":216000,"video_input":"none"}},{"label":"720p · 5s (+4s 输入视频)","facts":{"resolution":"720p","tokens":194400,"video_input":"video"}}]},{"model_name":"gpt-image-2.5-sunburst-low","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.03,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.1-pro-preview-customtools","vendor_id":6,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Gemini Vertex","Gemini Premium"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gpt-5.4","description":"GPT-5.4 将我们近期在推理、编码和智能体工作流程方面取得的最佳进展整合到一个前沿模型中。它融合了 GPT-5.3-Codex 业界领先的编码能力， 同时改进了模型在各种工具、软件环境以及涉及电子表格、演示文稿和文档的专业任务中的运行方式。最终，该模型能够准确、高效地完成复杂的实际工作，以更少的反复沟通交付您所需的结果。","tags":"对话,多模态,工具,推理","vendor_id":2,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Azure High","default","Azure All","Codex Pro","Codex Plus","Codex Team"],"supported_endpoint_types":["openai","openai responses"]},{"model_name":"claude-opus-4-20250514","description":"Claude Opus 4 (20250514) 是 Anthropic 推出的第四代顶级旗舰模型，被公认为 2025 年上半年全球最强的编程与科研模型。它是 Claude 4 家族中智能水平最高、推理深度最深的成员。该模型引入了“扩展思维 (Extended Thinking)”模式，支持长达数小时的自主任务执行。它在 SWE-bench 编程测试中取得了 72.5% 的惊人成绩，能够处理跨越成千上万个文件的复杂架构重构，并保持完美的逻辑一致性。Opus 4 专门为那些需要极高精确度、长程规划和复杂工具调用的“AI 代理 (AI Agents)”场景而设计。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":7.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Mix"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-5.5","vendor_id":2,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["Codex Team","Codex Plus","Azure High","default","Codex Pro"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-6-sol","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Azure High","default","Codex Pro","Codex Plus","Codex Team"],"supported_endpoint_types":["openai"]},{"model_name":"minimax-m2.5","tags":"对话,多模态,推理,工具","vendor_id":20,"quota_type":0,"model_ratio":1.05,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-3.5-turbo","vendor_id":2,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default","Azure All","Azure High"],"supported_endpoint_types":["openai"]},{"model_name":"doubao-seedance-1-0-pro-fast-251015","vendor_id":23,"quota_type":1,"model_ratio":0,"model_price":0.08,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-image-2.5-sunburst","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.075,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"deepseek-r1-0528","description":"deepseek-r1-0528 是 DeepSeek 专为解决复杂逻辑、数学、编程及长程规划任务而设计的推理模型 API 别名。与追求对话流畅度的 deepseek-chat 不同，该模型在输出最终答案前，会强制生成一段名为 Chain of Thought (CoT，思维链) 的推理过程。在 2026 年初，该接口主要指向经过大规模强化学习（RL）迭代后的 DeepSeek-R1 旗舰版本。它不仅在 AIME 和 Codeforces 等硬核测试中对标 OpenAI o1/o3，还引入了“自我反思”机制，能够识别并修正推理过程中的错误，是构建自动化研发、科学计算及高复杂性 Agent 的核心大脑。","tags":"对话,多模态,工具","vendor_id":4,"quota_type":0,"model_ratio":2,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"claude-haiku-4-5-20251001","description":"Claude Haiku 4.5 是 Anthropic 开发的最快、最经济高效的小型模型。作为 4.5 系列的入门级产品，它实现了显著的跨代跨越：其编码能力和 Agent 任务处理能力已能比肩半年前的顶级模型 Claude 3.5 Sonnet。该模型专门为高并发、低延迟的应用场景设计，如实时翻译、代码子代理和大规模金融数据分析。它是 Haiku 系列中首个原生支持“思维链 (CoT)”和“计算机操作 (Computer Use)”的版本，极大提升了在处理复杂逻辑时的准确率。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":0.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Kiro Business","Claude Code Max","Claude Mix","Claude Code","Claude Aws Code","default","Claude Vertex"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"doubao-seedream-4-5","vendor_id":23,"quota_type":1,"model_ratio":0,"model_price":0.05,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-image-2.5-flare","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.075,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"grok-4-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.005,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"grok-4.5-high-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.02,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-2.5-flash-image","vendor_id":6,"quota_type":1,"model_ratio":0,"model_price":0.07,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"text-embedding-3-large","quota_type":0,"model_ratio":0.065,"model_price":0,"owner_by":"","completion_ratio":0,"enable_groups":["Azure High"],"supported_endpoint_types":["openai"]},{"model_name":"deepseek-v3","description":"拥有 671B 总参数（37B 激活）的顶尖 MoE 模型。采用 MLA 架构和无辅助损失负载均衡技术。支持“思考/非思考”混合模式，在编程、数学和智能体任务中表现卓越，是 2026 年市场上性价比最高的旗舰模型。","tags":"对话,多模态,推理,工具","vendor_id":4,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"deepseek-v4-flash","vendor_id":4,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":3,"cache_ratio":0.1,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"o3-2025-04-16","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.1-flash-lite-preview","description":"Gemini 3.1 Flash-Lite Preview 在智能方面位列领先模型之列，但与其他同价位模型相比价格略高。它的速度也相当快，但生成的信息非常冗长。该模型支持文本和图像输入，输出文本，并拥有 100 万个标记的上下文窗口。\n\nGemini 3.1 Flash-Lite Preview 在人工智能分析智能指数 (AIEI) 中得分 34 分，远高于同类模型的平均水平（平均分为 19 分）。在评估智能指数时，它生成了 5300 万个标记，与平均 2000 万个标记相比，信息量非常大。","tags":"对话,多模态,推理,工具","vendor_id":6,"quota_type":0,"model_ratio":0.125,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Gemini AI Studio","default","Gemini Cli","Gemini Antigravity"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gpt-5.5-openai-compact","vendor_id":2,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Codex Pro","default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-image-2.5-low","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.03,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"kimi-k2.6","vendor_id":5,"quota_type":0,"model_ratio":3.25,"model_price":0,"owner_by":"","completion_ratio":4.153846153846,"cache_ratio":0.169230769231,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3-flash","vendor_id":6,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":6,"enable_groups":["Gemini Cli","Gemini Antigravity","default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"kimi-k2.7-code","vendor_id":5,"quota_type":0,"model_ratio":3.25,"model_price":0,"owner_by":"","completion_ratio":4.153846153846,"cache_ratio":0.2,"enable_groups":["Chinese Model","default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4.1-nano-2025-04-14","vendor_id":2,"quota_type":0,"model_ratio":0.05,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4o-2024-08-06","vendor_id":2,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.5,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"grok-4.3-fast-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.01,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"deepseek-v3.2","description":"DeepSeek-V3.2 是一款兼具高计算效率和出色推理能力及代理性能的模型。DeepSeek 的方法基于三项关键技术突破：DeepSeek 稀疏注意力 (DSA)、可扩缩的强化学习框架和大规模智能体任务合成流水线。","tags":"对话","vendor_id":4,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":1.5,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5","description":"OpenAI 的第五代旗舰集成模型。它首次在底层架构上融合了 GPT 系列（大语言模型）与 o 系列（推理模型），实现了“原生思维链”与“全能多模态”的统一。模型具备自动路由机制，能根据任务复杂度自主决定是否开启深度思考模式。","tags":"对话,推理,工具,识图,多模态","vendor_id":2,"quota_type":0,"model_ratio":0.625,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default","Azure All","Azure High"],"supported_endpoint_types":["openai","openai responses"]},{"model_name":"grok-420-agents-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.0075,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.3-codex-spark","description":"作为一款针对快速推理优化的高性能轻量级模型，在评估智能体软件工程能力的两个基准测试 SWE-Bench Pro 和 Terminal-Bench 2.0 中，GPT‑5.3‑Codex‑Spark 在完成任务时展现出强劲的性能，其所需时间仅为 GPT‑5.3‑Codex 的一小部分。","tags":"对话","vendor_id":2,"quota_type":0,"model_ratio":0.875,"model_price":0,"owner_by":"","completion_ratio":8,"enable_groups":["default","Codex Pro"],"supported_endpoint_types":["openai","openai responses"]},{"model_name":"bge-m3","quota_type":0,"model_ratio":0.03,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"bge-reranker-v2-m3","quota_type":0,"model_ratio":0.03,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4.1-mini-2025-04-14","vendor_id":2,"quota_type":0,"model_ratio":0.2,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-embedding-001","vendor_id":6,"quota_type":0,"model_ratio":0.075,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["Gemini Premium","Gemini AI Studio","Gemini Vertex"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gpt-6-astra","vendor_id":2,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Codex Pro","Codex Team","Codex Plus","Azure High","default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.6-luna","vendor_id":2,"quota_type":0,"model_ratio":0.5,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-embedding-2-preview","vendor_id":6,"quota_type":0,"model_ratio":0.15,"model_price":0,"owner_by":"","completion_ratio":3,"enable_groups":["Gemini Vertex","Gemini Premium"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gemini-embedding-2","vendor_id":6,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["Gemini AI Studio"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"grok-420-thinking-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.0075,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4o-mini","vendor_id":2,"quota_type":0,"model_ratio":0.075,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.5,"enable_groups":["Azure High","default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.6-sol","vendor_id":2,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default","Codex Pro","Codex Team","Codex Plus","Azure High"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-6.1-sol","vendor_id":2,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default","Codex Plus","Codex Pro","Codex Team"],"supported_endpoint_types":["openai"]},{"model_name":"claude-opus-4-8","vendor_id":1,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Vertex","Kiro Business","Claude Code Max","Claude Mix","Claude Code","Claude Aws Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gemini-2.5-pro-preview-tts","vendor_id":6,"quota_type":0,"model_ratio":0.5,"model_price":0,"owner_by":"","completion_ratio":20,"cache_ratio":0.1,"enable_groups":["Gemini Vertex","Gemini Premium"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"grok-4-deepsearch-c","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.005,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3-pro-image-preview","vendor_id":6,"quota_type":1,"model_ratio":0,"model_price":0.25,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"deepseek-v3-0324","description":"拥有 671B 总参数（37B 激活）的顶尖 MoE 模型。采用 MLA 架构和无辅助损失负载均衡技术。支持“思考/非思考”混合模式，在编程、数学和智能体任务中表现卓越，是 2026 年市场上性价比最高的旗舰模型。","tags":"对话,多模态,推理,工具","vendor_id":4,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4.1-2025-04-14","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"deepseek-v4-pro","vendor_id":4,"quota_type":0,"model_ratio":4.5,"model_price":0,"owner_by":"","completion_ratio":3,"cache_ratio":0.033333333333,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"grok-imagine-image-pro","vendor_id":24,"quota_type":1,"model_ratio":0,"model_price":0.065,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-2.5-pro","description":"Gemini 2.5 系列的旗舰级多模态模型。它采用了 Adaptive Thinking（自适应推理） 架构，能够根据问题难度动态调整思考深度。该模型在处理长达 100 万 token 的复杂代码库、财务报表和法律合同方面表现出卓越的稳定性，被广泛视为企业自动化、复杂 Agent 规划及深度研究任务的行业标准。","tags":"对话,多模态,推理,工具","vendor_id":6,"quota_type":0,"model_ratio":0.625,"model_price":0,"owner_by":"","completion_ratio":8,"enable_groups":["Gemini AI Studio","default","Gemini Antigravity","Gemini Cli","Gemini Vertex","Gemini Premium"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gemini-3.1-flash-lite-image","vendor_id":6,"quota_type":1,"model_ratio":0,"model_price":0.07,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gpt-6-luna","vendor_id":2,"quota_type":0,"model_ratio":0.05,"model_price":0,"owner_by":"","completion_ratio":5,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"deepseek-chat","description":"deepseek-chat 是 DeepSeek 官方 API 的动态指向别名。在 2026 年 1 月的当前版本中，它后端接入的是 DeepSeek-V3.2。该模型是 DeepSeek 家族中的全能旗舰，采用 671B 总参数的混合专家（MoE）架构，推理时激活约 37B 参数。与专注推理的 deepseek-reasoner (R1) 不同，deepseek-chat 默认运行于 Non-Thinking（非思维模式），追求极高的响应速度和流畅的对话体验。它不仅具备顶级的中文创作和通识能力，还通过集成 DSA (DeepSeek Sparse Attention) 技术，实现了在 101 万 超长上下文下的高效信息处理，是目前全球性价比最高的生产级 API 之一。","tags":"对话,多模态,工具","vendor_id":4,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.1,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-image-2","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.075,"owner_by":"","completion_ratio":0,"enable_groups":["OpenAI Image Official;","default"],"supported_endpoint_types":["openai"]},{"model_name":"kimi-k2-thinking","vendor_id":5,"quota_type":0,"model_ratio":1.8,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["Chinese Model","default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4.1","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":2,"cache_ratio":0.25,"enable_groups":["Azure All","Azure High","default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.5-flash","vendor_id":6,"quota_type":0,"model_ratio":0.75,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Gemini Premium","Gemini AI Studio","default","Gemini Antigravity","Gemini Vertex"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gemini-3-pro-preview","description":"Google Gemini 3 系列中最智能、最全能的旗舰模型。它专为超复杂任务设计，拥有顶级的多模态推理、博士级 STEM 问题解决能力以及卓越的自主代理（Agentic）性能。在 2025 年末的评测中，它是首个在 LMArena 榜单上突破 1500 Elo 分数的模型。","tags":"对话,多模态,推理,工具,废弃","vendor_id":6,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.25,"enable_groups":["Gemini Antigravity","Gemini Cli","Gemini Vertex","Gemini Premium","default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gpt-image-2.5-flare-low","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.03,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"glm-5.2","vendor_id":3,"quota_type":0,"model_ratio":4,"model_price":0,"owner_by":"","completion_ratio":3.5,"cache_ratio":0.25,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.1-pro-preview","description":"Gemini 3.1 Pro 是我们最先进的 Gemini 推理模型，能够解决复杂问题。凭借其 100 万个标记的上下文窗口，Gemini 3.1 Pro 可以理解来自不同信息源（包括文本、音频、图像、视频、PDF，甚至整个代码库）的海量数据集和具有挑战性的问题。有关使用最新 Gemini 模型的更多信息，请参阅 Gemini 3 入门指南。质量改进 Gemini 3.1 Pro 包含多项质量改进：改进的软件工程和代理能力：改进了软件工程行为和可用性，并在金融和电子表格应用程序等领域改进了代理功能。改进的标记效率和思维：在各种用例中实现更高效的思维。扩展的思维级别：引入了 MEDIUM 作为 thinking_level 参数，从而提供更多选项来优化成本、性能和速度之间的权衡。","tags":"对话,多模态,推理,工具","vendor_id":6,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Gemini AI Studio","default","Gemini Cli","Gemini Antigravity","Gemini Vertex","Gemini Premium"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gemini-2.5-flash","description":"Gemini 2.5 家族的“性价比之王”。它是首个将高吞吐量、低延迟特性与**原生推理（Thinking）**能力相结合的轻量级模型。它不仅能处理百万级上下文，还能通过分配“思维预算”来解决原本只有 Pro 模型才能处理的复杂逻辑任务。","tags":"对话,多模态,推理,工具","vendor_id":6,"quota_type":0,"model_ratio":0.15,"model_price":0,"owner_by":"","completion_ratio":8.333333333333334,"enable_groups":["default","Gemini Antigravity","Gemini Cli","Gemini Vertex","Gemini Premium"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"claude-opus-5","vendor_id":1,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Kiro Business","Claude Code Max","Claude Mix","Claude Aws Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-sonnet-4-20250514","description":"Claude Sonnet 4 是 Anthropic 4.0 系列的核心型号，旨在接替广受好评的 Claude 3.5 Sonnet。它引入了革命性的“混合推理”架构，允许模型在“即时响应”和“深度思考（Extended Thinking）”之间根据任务复杂度动态切换。Sonnet 4 在编程能力、指令遵循和长文本记忆力方面实现了跨越式提升，其在 SWE-bench 编程基准测试中达到了 72.7% 的历史高位。它不仅能处理极其复杂的代码重构，还能在长达 100 万 token 的超大上下文中实现近乎完美的记忆调取，是目前企业级 AI 代理和专业开发者的主流生产力工具。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Mix","Claude Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-4o-2024-05-13","vendor_id":2,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":3,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"doubao-seedance-2-0-fast-260128","vendor_id":23,"quota_type":0,"model_ratio":9.25,"model_price":0,"owner_by":"","completion_ratio":0.054054054054,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_usage_schema":{"resolution":{"enum":["480p","720p","1080p","4k"],"description":{"en":"Output video resolution","zh":"输出视频分辨率"},"enumLabels":{"1080p":{"en":"1080p","zh":"1080p"},"480p":{"en":"480p","zh":"480p"},"4k":{"en":"4k","zh":"4k"},"720p":{"en":"720p","zh":"720p"}}},"tokens":{"type":"number","unit":"token","description":{"en":"Billing token unit price","zh":"计费 Token 单价"}},"video_input":{"enum":["none","video"],"description":{"en":"Reference video input","zh":"参考视频输入"},"enumLabels":{"none":{"en":"No reference video","zh":"无参考视频"},"video":{"en":"With reference video","zh":"有参考视频"}}}},"billing_usage_examples":[{"label":"480p · 5s","facts":{"resolution":"480p","tokens":48038,"video_input":"none"}},{"label":"720p · 5s","facts":{"resolution":"720p","tokens":108000,"video_input":"none"}},{"label":"1080p · 5s","facts":{"resolution":"1080p","tokens":243000,"video_input":"none"}},{"label":"4k · 5s","facts":{"resolution":"4k","tokens":972000,"video_input":"none"}},{"label":"720p · 10s","facts":{"resolution":"720p","tokens":216000,"video_input":"none"}},{"label":"720p · 5s (+4s 输入视频)","facts":{"resolution":"720p","tokens":194400,"video_input":"video"}}]},{"model_name":"claude-opus-4-6-thinking","description":"Claude Opus 4.6 是 Anthropic 于 2026 年 2 月推出的顶奢旗舰模型，专为“端到端企业自动化”设计。它不仅是 Opus 4.5 的直接升级版，更在长文本处理上实现了量级突破。Opus 4.6 的核心竞争力在于其近乎人类的“自主权”，它能独立完成从“信息检索”到“深度分析”再到“最终交付物生成（如 PPT 或 Excel）”的全流程。该模型在网络安全领域表现惊人，无需特殊工具即可发现代码中隐藏数十年的 0-day 漏洞。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude Vertex","Kiro Business","Claude Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-opus-4-5-20251101","description":"Claude 4.5 Opus 是 Anthropic 目前功能最强大的模型。它在处理极其复杂的任务（如尖端科学研究、复杂的编码架构设计和深层次的战略分析）方面表现卓越。相比之前的版本，它显著提升了逻辑推理的连贯性，并大幅减少了幻觉。Opus 版本特别强化了对多模态输入的跨表分析能力，能够同时理解并关联数千页文档、代码库以及高清视频流中的细节。它还引入了更加细粒度的“宪法 AI”约束，使其在保持强大能力的同时，更加安全且符合人类价值观。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Kiro Business","Claude Code Max","Claude Mix","Claude Code","Claude Aws Code","default","Claude Vertex"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-fable-5","vendor_id":1,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Claude Code Max","default","Claude Mix","Claude Code","Claude Aws Code"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-image-2-low","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.03,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"qwen3-embedding-8b","vendor_id":21,"quota_type":0,"model_ratio":0.125,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","Chinese Model"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-3.5-turbo-16k","vendor_id":2,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default","Azure All"],"supported_endpoint_types":["openai"]},{"model_name":"claude-sonnet-4-6-thinking","description":"Sonnet 4.6 是 Anthropic 迄今为止功能最强大的 Sonnet 系列产品，在编码、代理和专业工作方面均展现出卓越的性能。它尤其擅长迭代开发、复杂代码库导航、端到端项目管理（含内存管理）、文档创建以及在 Web 质量保证和工作流自动化方面的出色计算机应用。","tags":"对话,多模态,推理,工具","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Claude Vertex","Kiro Business","Claude Code","default"],"supported_endpoint_types":["anthropic","openai"]}],"group_ratio":{"Azure All":0.7,"Azure High":1.8,"Chinese Model":0.5,"Chinese Model Cheap":0.3,"Claude Aws Code":0.95,"Claude Code":0.95,"Claude Code Max":2.5,"Claude Mix":0.6,"Claude Official":5.5,"Claude Vertex":0.95,"Codex Plus":0.25,"Codex Pro":0.35,"Codex Team":0.35,"Gemini AI Studio":2.5,"Gemini Antigravity":0.4,"Gemini Cli":0.5,"Gemini Premium":3.5,"Gemini Vertex":2.5,"Grok Reverse":0.3,"Kiro Business":0.7,"OpenAI Image Official;":1.7,"default":1},"pricing_version":"a42d372ccf0b5dd13ecf71203521f9d2","success":true,"supported_endpoint":{"anthropic":{"path":"/v1/messages","method":"POST"},"gemini":{"path":"/v1beta/models/{model}:generateContent","method":"POST"},"openai":{"path":"/v1/chat/completions","method":"POST"},"openai responses":{"path":"/v1/responses","method":"POST"}},"usable_group":{"Azure All":"Azure 分组","Azure High":"Azure High分组","Chinese Model":"国产专用模型","Chinese Model Cheap":"国产特价模型","Claude Aws Code":"","Claude Code":"Claude Code分组","Claude Code Max":"Claude Code Max 200刀分组","Claude Mix":"Claude 逆向混合分组","Claude Official":"Claude 纯官方分组","Claude Vertex":"Claude Vertex分组","Codex Plus":"Codex Plus分组","Codex Pro":"","Codex Team":"Codex 官方分组","Gemini AI Studio":"Gemini AI Studio分组","Gemini Antigravity":"Gemini Antigravity分组","Gemini Cli":"Gemini Cli分组","Gemini Premium":"Gemini 优质分组","Gemini Vertex":"Gemini Vertex分组","Grok Reverse":"Grok逆向分组","Kiro Business":"Kiro 企业版","OpenAI Image Official;":"GPT生图官方分组","default":"默认分组"},"vendors":[{"id":1,"name":"Anthropic","icon":"Claude.Color"},{"id":7,"name":"讯飞","icon":"Spark.Color"},{"id":8,"name":"Vivgrid"},{"id":10,"name":"OpenCode Zen","icon":"OpenCode"},{"id":13,"name":"Jiekou.AI"},{"id":15,"name":"302.AI"},{"id":16,"name":"OpenCode Go"},{"id":18,"name":"Zhipu AI Coding Plan"},{"id":2,"name":"OpenAI","icon":"OpenAI"},{"id":4,"name":"DeepSeek","icon":"DeepSeek.Color"},{"id":11,"name":"QiHang"},{"id":12,"name":"Helicone"},{"id":14,"name":"SAP AI Core"},{"id":19,"name":"Venice AI","icon":"VeniceAI"},{"id":21,"name":"阿里巴巴","icon":"Qwen.Color"},{"id":22,"name":"即梦","icon":"Jimeng.Color"},{"id":17,"name":"Vertex","icon":"VertexAI.Color"},{"id":24,"name":"xAI","icon":"XAI"},{"id":3,"name":"智谱","icon":"Zhipu.Color"},{"id":5,"name":"Moonshot","icon":"Moonshot"},{"id":6,"name":"Google","icon":"Gemini.Color"},{"id":9,"name":"Qiniu"},{"id":20,"name":"MiniMax","icon":"MiniMax"},{"id":23,"name":"字节跳动","icon":"Doubao.Color"}]}