{"auto_groups":[],"data":[{"model_name":"grok-4.5","description":"500K 上下文，支持图文输入，推理速度快。深度优化代码开发、代码审查与 AI 智能体工作流，支持工具调用，Token 效率高，适合大型项目开发、长文档分析、自动化任务搭建。","tags":"多模态,code,长文档处理","vendor_id":22,"quota_type":0,"model_ratio":0.6,"model_price":0,"owner_by":"","completion_ratio":3,"cache_ratio":0.25,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-response"],"billing_mode":"tiered_expr","billing_expr":"((len \u003c 200000)) ? (tier(\"standard\", p * 1.5 + c * 4 + cr * 0.3)) : (((len \u003e= 200000)) ? (tier(\"long_context\", p * 3 + c * 8 + cr * 0.6)) : (tier(\"base\", p * 1.5 + c * 4 + cr * 0.3)))","pricing_version":"5a90f2b86c08bd983a9a2e6d66c255f4eaef9c4bc934386d2b6ae84ef0ff1f1f"},{"model_name":"grok-4.6","description":"xAI 2026 前沿旗舰模型。500K 上下文，支持图文输入、工具调用。重点强化长链路 Agent 智能体、复杂代码工程、知识分析任务，评测追平 GPT‑5.6，推理档位可调，API 定价具备优势，适合自动化与开发场景","tags":"旗舰模型,code,多模态","vendor_id":22,"quota_type":0,"model_ratio":0.7,"model_price":0,"owner_by":"","completion_ratio":2.625,"cache_ratio":0.228571428571,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-response"],"billing_mode":"tiered_expr","billing_expr":"(len \u003c 200000) ? (tier(\"standard\", p * 1.5 + c * 4 + cr * 0.3)) : (len \u003e= 200000) ? (tier(\"long_context\", p * 3 + c * 8 + cr * 0.6)) : ((tier(\"base\", p * 1.5 + c * 4 + cr * 0.3)))"},{"model_name":"deepseek-v4.1-flash","description":"DeepSeek 最新轻量旗舰 MoE 模型，552B 总参数，采用全新 CED 非对称架构，输入仅激活 8B、输出 16B，推理成本大幅下降。原生支持图文多模态理解，1M 超长上下文，多项基准超越 V4-Pro，主打 \"高性能、低价格\" 的普惠路线。","tags":"高性价比,文本生成,深度思考","vendor_id":1,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", p * 0.5 + c * 2 + cr * 0.05)"},{"model_name":"minimax-m3","description":"MiniMax 旗舰模型，主打顶尖 Coding 能力 + 1M 超长上下文（MSA 稀疏注意力架构），原生支持图文 / 视频多模态理解。定位 \"一个模型搞定编程、长文档分析、Agent 工作流\"，同时推出低价 Token Plan 订阅制，主打让前沿模型进入开发者日常。","tags":"多模态,coding,旗舰模型","vendor_id":16,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"((len \u003c 250000)) ? (tier(\"standard\", p * 2.1 + c * 8.4 + cr * 0.42)) : (((len \u003e= 250000)) ? (tier(\"long_context\", p * 4.2 + c * 16.8 + cr * 0.84)) : (tier(\"base\", p * 2.1 + c * 8.4 + cr * 0.42)))"},{"model_name":"doubao-seed-2-0-code-preview-260215","description":"Doubao-Seed-2.0-Code 面向企业级编程需求优化，在 Seed 2.0 优秀的 Agent、VLM 能力基础上，特别增强了代码能力，不仅前端能力表现出众，也对企业常见的多语言编码需求做了特别优化，适合接入各种 AI 编程工具使用。","tags":"深度思考,code","vendor_id":9,"quota_type":0,"model_ratio":1.8,"model_price":0,"owner_by":"","completion_ratio":4.722222222222,"cache_ratio":0.194444444444,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-6.1-sol","description":"能力接近旗舰 GPT-6 Astra，标准 Token 价格仅为 Astra 的 1/5，缓存成本相比旧版 GPT-6 Sol 下降 50%。擅长真实代码库开发调试、复杂 PDF 长文档解析、科研任务、多步骤 Agent 自动化任务。","tags":"推理模型,Agent智能体,高性价比","vendor_id":8,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"((len \u003c 272000)) ? (tier(\"standard\", p * 0.9 + c * 4.5 + cr * 0.1)) : ((len \u003e= 272000)) ? (tier(\"long_context\", p * 1.8 + c * 7.5 + cr * 0.2)) : ((tier(\"base\", p * 0.9 + c * 4.5 + cr * 0.1)))"},{"model_name":"gpt-6-sol","description":"GPT-6 Sol 是 OpenAI GPT6 系列中端推理模型，继承 Astra 能力，支持图文输入，105 万超长上下文。擅长复杂推理、代码编写、Agent 自动化任务，自适应推理强度，相比前代大幅降价，兼顾性能与成本。","tags":"超长上下文,code,Agent能力","vendor_id":8,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"((len \u003c 272000)) ? (tier(\"standard\", p * 0.9 + c * 4.5 + cr * 0.18)) : ((len \u003e= 272000)) ? (tier(\"long_context\", p * 1.8 + c * 7.5 + cr * 0.26)) : ((tier(\"base\", p * 0.9 + c * 4.5 + cr * 0.18)))"},{"model_name":"claude-haiku-4-5","description":"Claude Haiku 4.5，Anthropic 2025 年 10 月发布轻量高速模型，支持 200K 上下文。首代 Haiku 搭载扩展思考与 Computer Use，代码能力对标 Sonnet 4，速度更快、成本仅约其三分之一。主打低延迟实时对话、批量任务、轻量 Agent，适合高并发规模化部署。","tags":"高性价比,快速响应","vendor_id":19,"quota_type":0,"model_ratio":0.35,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.257142857143,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-opus-5","description":"Claude Opus 5 是 Anthropic 新一代高端模型，支持 100 万上下文窗口。性能接近旗舰 Fable 5，定价维持前代 Opus 4.8 水平。代码、复杂推理、长文档分析突出，支持多级推理强度调节，适合代码评审、大型项目分析、专业深度任务。","tags":"code,复杂推理,多模态","vendor_id":19,"quota_type":0,"model_ratio":1.3,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.115384615385,"create_cache_ratio":1.230769230769,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-5.6-terra","description":"GPT-5.6-Terra 是一款注重稳定性与通用能力的大模型，在文本理解、逻辑推理和多轮对话中表现均衡。适用于内容生成、知识问答及业务辅助等场景，同时兼顾响应速度与成本控制，适合作为企业级应用或中转网关中的基础模型使用。","tags":"稳定通用,code","vendor_id":8,"quota_type":0,"model_ratio":0.6,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.125,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"((len \u003c 272000)) ? (tier(\"standard\", p * 1.2 + c * 6 + cr * 0.15)) : (((len \u003e= 272000)) ? (tier(\"long_context\", p * 2.4 + c * 10 + cr * 0.3)) : (tier(\"base\", p * 1.2 + c * 6 + cr * 0.15)))"},{"model_name":"happyhorse-1.0-i2v","description":"HappyHorse-1.0-I2V支持图生视频，具备高度还原的动态画面生成能力，能够精准理解文本语义，输出流畅自然、细节丰富的高质量视频。换算成秒是1.5元/秒","tags":"视频生成","vendor_id":2,"quota_type":0,"model_ratio":0.00005,"model_price":0,"owner_by":"","completion_ratio":500000,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.7-flash","description":"Gemini‑3.7‑Flash 是谷歌 2026 年 8 月发布的高性价比多模态主力模型抖音百科。拥有 1M 上下文、64K 输出，支持图文音视频输入，强化代码与智能体能力，可自定义推理模式，兼顾速度、成本与复杂任务表现，适配开发与企业工作流","tags":"多模态,code,复杂任务","vendor_id":20,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.15,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"deepseek-v4-pro","description":"旗舰级 MoE 大模型，总参1.6T、激活 49B，原生支持百万级超长上下文。依托海量高质量训练数据，具备顶尖数学逻辑、复杂推理、专业代码与长文本深度解析能力，适配高阶科研、复杂办公、深度智能代理等高难度场景。","tags":"深度思考,旗舰全能,code","vendor_id":1,"quota_type":0,"model_ratio":1.6,"model_price":0,"owner_by":"","completion_ratio":3.125,"cache_ratio":0.09375,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"glm-5.3-flash","description":"GLM‑5.3‑Flash 是智谱开源的 GLM‑5 系列原生多模态 MoE 模型。总参 320B，推理仅激活 18B，支持 1M 上下文，混合注意力架构，MIT 协议开源。编程、Agent 能力突出，性价比极高，兼顾图像理解，主打 API 高频调用与本地部署。","tags":"高新价比,高性能","vendor_id":3,"quota_type":0,"model_ratio":0.28,"model_price":0,"owner_by":"","completion_ratio":3.071428571429,"cache_ratio":0.392857142857,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"deepseek-flash","description":"DeepSeek V4.1 Flash 是 DeepSeek 发布的最小尺寸旗舰模型，原生支持多模态视觉理解，API 模型名为 deepseek-flash。该模型为 552B 参数 MoE 架构，输入激活仅 8B，输出激活 16B，推理成本显著降低","tags":"多模态,旗舰模型","vendor_id":1,"quota_type":0,"model_ratio":0.4,"model_price":0,"owner_by":"","completion_ratio":2.25,"cache_ratio":0.25,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", p * 0.5 + c * 2 + cr * 0.05)"},{"model_name":"gemini-3.5-flash","description":"Gemini 3.5 Flash 是 Google 推出的高性价比多模态模型，主打低延迟与高吞吐，适合实时对话、内容生成与轻量推理任务。相比 Pro 系列，Flash 在保证基础智能能力的同时显著降低成本，支持文本、图像等多模态输入，并具备较强的代码生成与理解能力，广泛应用于聊天机器人、自动化工具与前端交互等场景。","tags":"多模态,code,高性价比","vendor_id":20,"quota_type":0,"model_ratio":0.75,"model_price":0,"owner_by":"","completion_ratio":6.666666666667,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gpt-6-astra","description":"GPT‑6‑Astra 是 OpenAI2026 年 9 月发布的新一代旗舰大模型新华网。拥有 105 万上下文窗口，强化深度推理、计算机操作、编程与网络安全能力，擅长复杂多步骤智能体任务，支持图文输入，面向专业科研、工程与企业工作场景，逐步向订阅及 API 用户开放Azure AI F...","tags":"旗舰模型,code,多模态","vendor_id":8,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5.2,"cache_ratio":0.09,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"((len \u003c 272000)) ? (tier(\"standard\", p * 4 + c * 20.5 + cr * 0.4 + cc * 5)) : (((len \u003e= 272000)) ? (tier(\"long_context\", p * 7 + c * 31.5 + cr * 0.8)) : (tier(\"base\", p * 4 + c * 20.5 + cr * 0.42)))"},{"model_name":"grok-4.7","description":"Grok 4.7 是 SpaceXAI 推出的大模型，主打编程与专业知识处理。更大基座模型，强化长任务持续处理、长上下文与结果自检能力，安全防护升级。官方称速度为竞品两倍、成本更低，开放 API 接入 Cursor 等平台。","tags":"长上下文,code,高速推理","vendor_id":22,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-response"],"billing_mode":"tiered_expr","billing_expr":"(len \u003c 200000) ? (tier(\"standard\", p * 1.5 + c * 4 + cr * 0.35)) : ((len \u003e= 200000)) ? (tier(\"long_context\", p * 1.5 + c * 4 + cr * 0.35)) : ((tier(\"base\", p * 1.5 + c * 4 + cr * 0.35)))"},{"model_name":"claude-sonnet-5-5","description":"是 Anthropic 在 2026-09-28 发布的 Sonnet 系列主力升级模型，为 Claude 5.5 家族成员，上代 Sonnet 5 的直接迭代版本Anthropic。相比 Sonnet 5：推理速度提升 30%+，同等任务 Token 消耗减少，整体任务成本最高下降 30%**，综合能力大幅逼近旗舰 Opus 5.5，长周期 Agent 任务表现优于 GPT-6 Sol","tags":"Agent智能体,code,高性价比","vendor_id":19,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", p * 1.3 + c * 6 + cr * 0.2)"},{"model_name":"qwen3.6-plus","description":"Qwen3.6原生视觉语言系列Plus模型，展现出与当前顶尖前沿模型相媲美的卓越性能，模型效果相较3.5系列显著提升。模型在Agentic coding、前端编程、Vibe coding等代码能力、多模态万物识别、OCR、物体定位等能力上显著增强。","tags":"深度思考,视觉理解,文本生成","vendor_id":2,"quota_type":0,"model_ratio":0.75,"model_price":0,"owner_by":"","completion_ratio":4.8,"cache_ratio":0.2,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-4-argon","description":"谷歌 Gemini4 旗舰正式版，百万 token 上下文，原生多模态，擅长编码、科研、法律金融、网络安全，支持可调思考层级。","tags":"旗舰模型,代码推理,多模态","vendor_id":20,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", p * 3.5 + c * 18 + cr * 0.22)"},{"model_name":"gemini-3.6-flash","description":"Gemini 3.6 Flash是Google DeepMind推出的通用主力多模态大模型，作为3.5 Flash迭代版本，支持文本、图像、音视频输入，拥有百万级上下文窗口。模型优化代码生成、工具调用与智能体任务，显著减少冗余输出，Token消耗相比前代降低约17%，推理速度更快、调用成本下调，适合批量文档处理、自动化Agent、开发运维等规模化商用场景。","tags":"多模态,code","vendor_id":20,"quota_type":0,"model_ratio":1.1,"model_price":0,"owner_by":"","completion_ratio":5.454545454545,"cache_ratio":0.145454545455,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"gpt-image-2.5","description":"gpt-image-2.5 系列*，画质 \u003e gpt-image-2，延迟直接砍半，适合批量 / 快速迭代","tags":"画质好,速度快,生图","vendor_id":8,"quota_type":1,"model_ratio":0,"model_price":0.1,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["image-generation","openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", fixed(0.1))"},{"model_name":"claude-sonnet-4-6","description":"Claude Sonnet 4.6，Anthropic 2026 年 2 月发布中端主力模型，支持百万 token 上下文。代码、智能体、计算机操作能力大幅增强，性能贴近 Opus 4.6，成本更低。擅长批量开发、文档分析、自动化 Agent 任务，兼顾速度与推理质量，适合规模化商用场景。","tags":"code,多模态","vendor_id":19,"quota_type":0,"model_ratio":0.8,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.125,"create_cache_ratio":1.1875,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-fable-5-1","description":"Anthropic 顶尖通用大模型，1M 上下文，最大输出 128K。擅长长周期 Agent 任务、复杂编码、多步骤科研与文档处理，推理能力强，缓存读取费用大幅降低，适合高难度复杂工作流，有安全防护限制高危领域调用","tags":"复杂编码,旗舰模型,code","vendor_id":19,"quota_type":0,"model_ratio":4,"model_price":0,"owner_by":"","completion_ratio":3.75,"cache_ratio":0.00625,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-haiku-5-5","description":"Claude Haiku 5.5 是一款高速、低成本的 AI 模型，擅长日常编程、代码分析、图像理解与指令执行，兼顾响应速度和实用性。","tags":"快速响应,高性价比,code","vendor_id":19,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"],"billing_mode":"tiered_expr","billing_expr":"((len \u003c 100000)) ? (tier(\"standard\", p * 0.3 + c * 1.5 + cr * 0.06 + cc * 0.4)) : (((len \u003e= 100000)) ? (tier(\"long_context\", p * 1.5 + c * 8.5 + cr * 0.3 + cc * 2)) : (tier(\"base\", p * 0.3 + c * 1.5 + cr * 0.06 + cc * 0.4)))"},{"model_name":"deepseek-v4-flash","description":"已全面转发到DeepSeek V4.1 Flash。高效轻量化MoE模型，总参284B，激活13B，原生支持百万超长上下文能力。推理速度快、延迟低、调用成本低廉，综合能力均衡，主打高并发、轻量化任务，适合日常对话、内容创作、基础 RAG、批量文案处理等普惠刚需场景。","tags":"文本生成,深度思考","vendor_id":1,"quota_type":0,"model_ratio":0.4,"model_price":0,"owner_by":"","completion_ratio":2.25,"cache_ratio":0.1875,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", p * 0.5 + c * 2 + cr * 0.05)"},{"model_name":"claude-opus-4-7","description":"Claude Opus 4.7 是 Anthropic 2026 年 4 月推出的旗舰开放模型。擅长复杂代码、长周期 Agent 任务，指令遵循更强；支持最高 2576px 高清图像解析，视觉细节识别大幅提升，适配图表、截图精读，拥有百万级上下文，适合高难度专业推理与多模态复杂工作。","tags":"code,多模态,旗舰全能","vendor_id":19,"quota_type":0,"model_ratio":1.3,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.115384615385,"create_cache_ratio":1.230769230769,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-opus-4-8","description":"Claude Opus 4.8 是一款高端旗舰级大模型，专注于复杂推理、长文本理解与高质量代码生成能力。在多轮对话和深度分析场景中表现稳定，适用于高要求的开发辅助、系统设计及企业级应用。同时具备较强的上下文处理能力，在性能与可靠性方面处于领先水平。","tags":"文本理解,code","vendor_id":19,"quota_type":0,"model_ratio":1.3,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.115384615385,"create_cache_ratio":1.230769230769,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-5.5","description":"GPT-5.5 是一款性能进一步提升的通用大模型，在代码生成、复杂推理和多轮对话方面表现更强。相比前代版本，其在稳定性、上下文理解及响应质量上有明显优化，适用于开发辅助、内容生成、数据分析等多种场景，同时在性能与成本之间保持良好平衡，适合作为中高阶应用的核心模型。","tags":"通用大模型,code,多模态","vendor_id":8,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":5.6,"cache_ratio":0.096,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"((p \u003c 272000)) ? (tier(\"standard\", p * 2.5 + c * 13 + cr * 0.24)) : (((len \u003e= 272000)) ? (tier(\"long_context\", p * 5 + c * 25 + cr * 0.4)) : (tier(\"base\", p * 2.5 + c * 13 + cr * 0.24)))"},{"model_name":"claude-opus-4-6","description":"Claude Opus 4.6 是 Anthropic2026 年 2 月发布的旗舰模型，支持100 万 token 超长上下文，强化 Agent 长任务、大型代码库分析与多步骤推理。搭载自适应思考机制，擅长复杂工程、深度文档研判；视觉能力稳健，适合企业复杂工作流、长期自主智能体开发。","tags":"code,多模态,复杂工作","vendor_id":19,"quota_type":0,"model_ratio":1.15,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.130434782609,"create_cache_ratio":1.247376311844,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gemini-3.1-pro-preview","description":"Gemini 3.1 Pro Preview 是 Google 推出的预览版旗舰推理模型，重点提升复杂推理、代码生成、Agent 工具调用及多步骤任务执行能力。模型具备更高的事实一致性、更优的 Token 效率，并支持最长 100 万 Token 上下文窗口，可处理超长文档、代码库及多模态输入，适用于软件开发、企业智能体、科研分析和复杂业务场景。","tags":"code,多模态,旗舰全能","vendor_id":20,"quota_type":0,"model_ratio":1.95,"model_price":0,"owner_by":"","completion_ratio":6.410256410256,"cache_ratio":0.102564102564,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"claude-fable-5","description":"Claude Fable 5 是一款偏向创意表达与叙事能力优化的大模型，在故事生成、文案创作和情感表达方面表现突出。其语言风格自然流畅，适合用于内容创作、品牌文案及交互式故事场景，同时兼顾一定的逻辑能力，能够在创意与实用之间取得良好平衡。","tags":"高新能,期间全能","vendor_id":19,"quota_type":0,"model_ratio":2.1,"model_price":0,"owner_by":"","completion_ratio":4.761904761905,"cache_ratio":0.119047619048,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", p * 4.2 + c * 20 + cr * 0.5)"},{"model_name":"kimi-k3.1-pro","description":"kimi-k3.1-pro 是月之暗面 Kimi K3.1 系列旗舰思考模型，3 万亿级参数，**1M token 超长上下文、原生多模态视觉理解**，强制开启推理思考，支持三档推理强度可调；重点强化代码开发、3D 空间建模、金融研究建模、长任务 Agent 能力，适配 Codex、Kimi Code 等编程代理、深度文档研读、复杂建模推演场景","tags":"旗舰模型,多模态,code","vendor_id":17,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", p * 4.5 + c * 26 + cr * 0.5)"},{"model_name":"gpt-6.1-terra","description":"GPT-6.1 Terra 是 OpenAI 高性价比办公模型，主打电脑操作、视觉推理，响应快，API 价格大幅下调，适合 AI Agent、自动化办公、图纸解析与编程，综合能力贴近旗舰。","tags":"高性价比,视觉推理,coding","vendor_id":8,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"((len \u003c 272000)) ? (tier(\"standard\", p * 0.65 + c * 3 + cr * 0.06)) : ((len \u003e= 272000)) ? (tier(\"long_context\", p * 1 + c * 5 + cr * 0.12)) : ((tier(\"base\", p * 0.65 + c * 3 + cr * 0.06)))"},{"model_name":"doubao-seed-2-0-pro-260215","description":"旗舰级全能通用模型，面向 Agent 时代的复杂推理与长链路任务执行场景。强调多模态理解、长上下文推理、结构化生成与工具增强执行。复杂指令与多约束执行能力突出，可稳定应对多步复杂规划、复杂图文推理、视频内容理解与高难度分析等场景。","tags":"推理模型,旗舰全能","vendor_id":9,"quota_type":0,"model_ratio":2,"model_price":0,"owner_by":"","completion_ratio":4.5,"cache_ratio":0.2,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.6-sol","description":"GPT-5.6-SOL 是一款高性能通用大模型，主打代码生成、复杂推理与多任务处理能力。相比基础版本，其在长上下文理解与稳定性上进一步优化，适合用于 AI 编程助手、自动化流程、数据分析及企业级应用场景。同时兼顾成本与响应速度，在多模型路由体系中具备较高性价比，是构建中转网关与开发工具链的理想选择。","tags":"code,高性能,多模态","vendor_id":8,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":5.6,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"(len \u003c 272000) ? (tier(\"standard\", p * 2 + c * 12 + cr * 0.2)) : (len \u003e= 272000) ? (tier(\"long_context\", p * 4 + c * 18 + cr * 0.4)) : ((tier(\"base\", p * 2 + c * 12 + cr * 0.2)))"},{"model_name":"gpt-6-luna","description":"GPT-6 Luna 是 OpenAI GPT6 系列高性价比轻量模型，105 万上下文，支持图文输入与工具调用。主打大批量高频任务，可调节推理强度，价格大幅降低，适合批量文本处理、日常对话与轻量 Agent 场景。","tags":"高性价比,多模态,百万上下文","vendor_id":8,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"((len \u003c 272000)) ? (tier(\"standard\", p * 0.6 + c * 2.2 + cr * 0.12)) : (((len \u003e= 272000)) ? (tier(\"long_context\", p * 1.2 + c * 4 + cr * 0.24)) : (tier(\"base\", p * 0.6 + c * 2.2 + cr * 0.12)))"},{"model_name":"qwen3.7-max","description":"Qwen3.7系列中规模最大、综合能力最强的Max模型，当前开放纯文本模型能力供体验。Qwen3.7是面向智能体时代的新一代旗舰模型，核心优势在于智能体能力的广度与深度：在编程、办公与生产力、长周期自主执行方面均能出色胜任各项任务。","tags":"旗舰全能,深度思考,code","vendor_id":2,"quota_type":0,"model_ratio":3,"model_price":0,"owner_by":"","completion_ratio":3,"cache_ratio":0.2,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-opus-5-5","description":"Anthropic 旗舰模型，百万 token 上下文，支持图文多模态。擅长 Agent 长周期编程、复杂专业推理与计算机操作，性能对标 Fable 5.1，相比 Opus 5 成本下降 40%，安全对齐更强，适合重度工程、文档分析任务。","tags":"旗舰推理,Agent编程,百万上下文","vendor_id":19,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", p * 2.2 + c * 11.4 + cr * 0.18)"},{"model_name":"glm-5.3","description":"依靠大规模后训练实现能力跃升。编程能力较上代提升 50%，长程软件工程、终端 Agent 表现突出，网络安全漏洞挖掘能力涌现，多项基准位居开源前列。支持 200K 上下文、128K 最大输出，适配代码开发、安全审计、工具调用场景，API 已开放，权重后续开源，适合开发与安全场景调用。","tags":"code,旗舰模型","vendor_id":3,"quota_type":0,"model_ratio":3.5,"model_price":0,"owner_by":"","completion_ratio":3.428571428571,"cache_ratio":0.214285714286,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", p * 5.3 + c * 17.2 + cr * 1)"},{"model_name":"qwen3.7-plus","description":"Qwen3.7系列中高性价比Plus模型，在强大文本能力的基础上全面升级了视觉-语言能力，同时保持了在编码、工具使用和生产力工作流方面的完整智能体能力。其核心特色为多模态交互混合智能体能力，能够感知真实世界场景、读取屏幕并操作 GUI、基于视觉参考生成代码、端到端导航移动应用。\n\n该模型版本功能等同于快照模型qwen3.7-plus-2026-05-26","tags":"文本生成,视觉理解,深度思考","vendor_id":2,"quota_type":0,"model_ratio":0.8,"model_price":0,"owner_by":"","completion_ratio":4.0625,"cache_ratio":0.25,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.8-flash","description":"Gemini‑3.8‑Flash，谷歌 2026‑09 发布，迭代自 3.7‑Flashstorage.go...。1M 上下文、64K 输出，全模态输入，强化代码、Agent 与多步推理，支持可调推理强度，速度成本不变，面向生产级智能体与开发场景","tags":"推理强,code,多模态","vendor_id":20,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.15,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"qwen3.8-max","description":"2.4万亿参数MoE旗舰，编程与办公能力全面跃升，可自主编程十数天交付完整项目。胜任法律、金融、设计等数百种专业任务，一次对话端到端交付生产级成果。原生视觉理解贯穿规划、执行与验证全流程，支持超长文档与长视频的深度语义解析。长程任务中自主规划与闭环迭代，持续进化。","tags":"深度思考,多模态,文本生成","vendor_id":2,"quota_type":0,"model_ratio":3,"model_price":0,"owner_by":"","completion_ratio":3,"cache_ratio":0.15,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"kimi-k3","description":"Kimi K3 是 Kimi 迄今能力最强的旗舰模型，拥有 2.8 万亿参数，基于 KDA 混合线性注意力机制（Kimi Delta Attention）和注意力残差（Attention Residuals）技术构建，原生支持视觉理解，并拥有 100 万 token 上下文窗口。它是全球首个开源的 3 万亿级别模型，面向长程编程、知识工作和推理等前沿智能场景而设计。","tags":"文本生成,旗舰模型","vendor_id":17,"quota_type":0,"model_ratio":8,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", p * 12 + c * 60 + cr * 1.2)"},{"model_name":"MiniMax-M2.7","description":"M2.7 能够自行构建复杂 Agent Harness，并基于 Agent Teams、复杂 Skills、Tool Search tool 等能力，完成高度复杂的生产力任务。","tags":"深度思考,文本生成","vendor_id":16,"quota_type":0,"model_ratio":0.91,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.236263736264,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"doubao-seed-1-8-251228","description":"Doubao-Seed-1.8 面向多模态 Agent 场景定向优化。Agent 能力上，Tool Use、复杂指令遵循等能力均大幅增强。多模态理解方面，视觉基础能力显著提升，可低帧率理解超长视频，视频运动理解、复杂空间理解及文档结构化解析能力也有所优化，还原生支持智能上下文管理，用户可配置上下文策略。","tags":"深度思考,视觉理解","vendor_id":9,"quota_type":0,"model_ratio":0.5,"model_price":0,"owner_by":"","completion_ratio":2,"cache_ratio":0.3,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-image-2","description":"OpenAI 2026 年 4 月发布旗舰图像生成模型（ChatGPT Images 2.0），搭载思考模式，指令遵循、空间逻辑大幅升级。中文等多语种文字渲染准确率极高，支持自定义画幅、最高 2K 输出，擅长海报、界面、图文排版，适配商业设计、对话式迭代修图。","tags":"图像生成,商业设计","vendor_id":8,"quota_type":1,"model_ratio":0,"model_price":0.08,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["image-generation","openai"],"billing_mode":"tiered_expr","billing_expr":"tier(\"base\", fixed(0.08))"},{"model_name":"qwen-image-2.0","description":"Qwen-Image-2.0系列加速版模型，实现了图片生成和图片编辑的融合；具备更专业的文字渲染1k token指令支持能力、更细腻的真实质感，细腻刻画写实场景、更强的语义遵循能力。加速版有效实现了模型效果和性能的最佳平衡。","tags":"图片生成","vendor_id":2,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["image-generation","openai"],"billing_usage_schema":{"image_count":{"type":"number","unit":"count","unitLabel":{"en":"image","fr":"image","ja":"枚","ru":"изображение","vi":"ảnh","zh":"张","zh-TW":"張"},"description":{"en":"Image generation unit price","zh":"图片生成单价"}}}},{"model_name":"doubao-seed-2-0-mini-260428","description":"量均衡型全模态模型，支持文本、图片、语音、视频四模态理解与跨模态推理。精简推理链路以提升响应效率和运行稳定性，强化多语种及小语种处理能力，适用于通用轻量化多模态解析场景。","tags":"深度思考,视觉理解","vendor_id":9,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.2,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.5-flash-lite","description":"Gemini 3.5 Flash Lite 是一款轻量级高效模型，主打低成本与极速响应，适合高频调用场景。具备基础的文本理解、简单推理及内容生成能力，可用于日常问答、轻量文案和自动化任务处理。在保证可用性的同时大幅降低算力消耗，适合作为中转网关或批量任务中的经济型选择。","tags":"文本理解,高性价比","vendor_id":20,"quota_type":0,"model_ratio":0.55,"model_price":0,"owner_by":"","completion_ratio":8.727272727273,"cache_ratio":0.109090909091,"enable_groups":["default"],"supported_endpoint_types":["gemini","openai"]},{"model_name":"claude-sonnet-5","description":"Claude Sonnet 5 是一款注重综合能力与稳定性的中高端大模型，在代码生成、文本理解和多轮对话方面表现优秀。相比轻量模型，其在复杂任务处理和逻辑推理上更为可靠，适用于开发辅助、内容创作及企业级应用场景，同时在性能与成本之间取得良好平衡。","tags":"code,文本理解,多模态","vendor_id":19,"quota_type":0,"model_ratio":0.6,"model_price":0,"owner_by":"","completion_ratio":4.666666666667,"cache_ratio":0.138039215687,"enable_groups":["default"],"supported_endpoint_types":["anthropic","openai"]}],"group_ratio":{"default":1},"pricing_version":"a42d372ccf0b5dd13ecf71203521f9d2","success":true,"supported_endpoint":{"anthropic":{"path":"/v1/messages","method":"POST"},"gemini":{"path":"/v1beta/models/{model}:generateContent","method":"POST"},"image-generation":{"path":"/v1/images/generations","method":"POST"},"openai":{"path":"/v1/chat/completions","method":"POST"},"openai-response":{"path":"/v1/responses","method":"POST"}},"usable_group":{"default":"默认分组"},"vendors":[{"id":1,"name":"DeepSeek","icon":"DeepSeek.Color"},{"id":2,"name":"阿里巴巴","icon":"Qwen.Color"},{"id":9,"name":"字节跳动","icon":"Doubao.Color"},{"id":16,"name":"MiniMax","icon":"Minimax.Color size={64}"},{"id":19,"name":"Anthropic","icon":"Claude.Color"},{"id":22,"name":"xAI","icon":"XAI"},{"id":3,"name":"智谱","icon":"Zhipu.Color"},{"id":7,"name":"腾讯","icon":"Hunyuan.Color"},{"id":8,"name":"OpenAI","icon":"OpenAI"},{"id":17,"name":"Kimi","icon":"Kimi.Color size={64} "},{"id":20,"name":"Google","icon":"Gemini.Color"}]}