{
  "updatedAt": "2026-04-08T21:38:04.947Z",
  "count": 13,
  "shortlistReady": true,
  "comparePath": "/compare",
  "items": [
    {
      "slug": "gpt-5-4-pro",
      "name": "GPT-5.4 Pro",
      "shortName": "GPT-5.4 Pro",
      "family": "GPT-5",
      "providerId": "openai",
      "officialUrl": "https://openai.com/api/",
      "docsUrl": "https://platform.openai.com/docs/models",
      "pricingUrl": "https://openai.com/api/pricing/",
      "description": {
        "en": "Top-tier generalist model with excellent reasoning depth, strong coding reliability, and mature agent tooling support.",
        "zh-TW": "頂級全能模型，具備出色的推理深度、穩定的程式能力，以及成熟的 agent 工具鏈支援。"
      },
      "modalities": [
        "text",
        "image",
        "tools"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "image",
          "file"
        ],
        "output": [
          "text",
          "json"
        ]
      },
      "contextWindow": "400K",
      "pricing": {
        "input": "$12 / 1M tok",
        "output": "$48 / 1M tok",
        "inputUsdPerMillion": "12",
        "outputUsdPerMillion": "48"
      },
      "speedCategory": "Balanced",
      "toolSupport": {
        "functionCalling": true,
        "structuredOutput": true,
        "webSearch": true,
        "remoteMcp": true,
        "codeExecution": true
      },
      "overallScore": 93,
      "scores": {
        "capability": 97,
        "useCaseFitness": 94,
        "costEfficiency": 71,
        "speed": 82,
        "reliability": 94,
        "agentReadiness": 97,
        "ecosystem": 98
      },
      "bestUseCases": [
        "coding",
        "research",
        "agent-automation"
      ],
      "worksWith": [
        "Codex CLI",
        "LangGraph",
        "MCP servers",
        "Playwright"
      ],
      "tags": [
        "flagship",
        "coding",
        "agents",
        "multimodal"
      ],
      "strengths": [
        {
          "en": "Very strong long-horizon task execution",
          "zh-TW": "長流程任務執行能力非常強"
        },
        {
          "en": "Deep ecosystem integration across tools and APIs",
          "zh-TW": "工具與 API 生態整合成熟"
        }
      ],
      "caveats": [
        {
          "en": "Premium pricing for heavy batch usage",
          "zh-TW": "大量批次使用時成本偏高"
        },
        {
          "en": "Fast enough, but not the cheapest latency pick",
          "zh-TW": "速度夠用，但不是最低延遲與最低成本選擇"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:gpt-5-4-pro",
      "status": "active",
      "displayName": "GPT-5.4 Pro",
      "summary": {
        "en": "Top-tier generalist model with excellent reasoning depth, strong coding reliability, and mature agent tooling support.",
        "zh-TW": "頂級全能模型，具備出色的推理深度、穩定的程式能力，以及成熟的 agent 工具鏈支援。"
      },
      "bestFor": [
        "coding",
        "research",
        "agent-automation"
      ],
      "sourceRefs": [
        {
          "id": "official-site:gpt-5-4-pro-official",
          "label": "GPT-5.4 Pro official",
          "url": "https://openai.com/api/",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:gpt-5-4-pro-docs",
          "label": "GPT-5.4 Pro docs",
          "url": "https://platform.openai.com/docs/models",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "pricing-page:gpt-5-4-pro-pricing",
          "label": "GPT-5.4 Pro pricing",
          "url": "https://openai.com/api/pricing/",
          "kind": "pricing-page",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "pricing"
          ]
        },
        {
          "id": "manual-review:gpt-5-4-pro-verdictlens-review",
          "label": "GPT-5.4 Pro VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "gpt-5-mini",
      "name": "GPT-5 Mini",
      "family": "GPT-5",
      "providerId": "openai",
      "officialUrl": "https://openai.com/api/",
      "docsUrl": "https://platform.openai.com/docs/models",
      "pricingUrl": "https://openai.com/api/pricing/",
      "description": {
        "en": "Lean, affordable OpenAI model tuned for responsive assistants, classification, and operational agent tasks.",
        "zh-TW": "輕量且價格友善的 OpenAI 模型，適合即時助理、分類與日常 agent 任務。"
      },
      "modalities": [
        "text",
        "image",
        "tools"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "image",
          "file"
        ],
        "output": [
          "text",
          "json"
        ]
      },
      "contextWindow": "200K",
      "pricing": {
        "input": "$1.1 / 1M tok",
        "output": "$4.4 / 1M tok",
        "inputUsdPerMillion": "1.1",
        "outputUsdPerMillion": "4.4"
      },
      "speedCategory": "Fast",
      "toolSupport": {
        "functionCalling": true,
        "structuredOutput": true,
        "webSearch": true,
        "remoteMcp": true,
        "codeExecution": true
      },
      "overallScore": 87,
      "scores": {
        "capability": 85,
        "useCaseFitness": 88,
        "costEfficiency": 94,
        "speed": 94,
        "reliability": 88,
        "agentReadiness": 91,
        "ecosystem": 96
      },
      "bestUseCases": [
        "agent-automation",
        "research"
      ],
      "worksWith": [
        "n8n",
        "Zapier AI Actions",
        "1Password CLI",
        "MCP servers"
      ],
      "tags": [
        "efficient",
        "assistant",
        "ops"
      ],
      "strengths": [
        {
          "en": "Excellent throughput-per-dollar",
          "zh-TW": "單位成本產出效率很高"
        },
        {
          "en": "Strong tool-calling ergonomics",
          "zh-TW": "工具呼叫體驗成熟穩定"
        }
      ],
      "caveats": [
        {
          "en": "Less depth for frontier-grade reasoning",
          "zh-TW": "面對最前沿推理任務時深度較不足"
        },
        {
          "en": "May need stronger prompt scaffolding for nuanced analysis",
          "zh-TW": "做細緻分析時需要更好的提示框架"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:gpt-5-mini",
      "status": "active",
      "displayName": "GPT-5 Mini",
      "summary": {
        "en": "Lean, affordable OpenAI model tuned for responsive assistants, classification, and operational agent tasks.",
        "zh-TW": "輕量且價格友善的 OpenAI 模型，適合即時助理、分類與日常 agent 任務。"
      },
      "bestFor": [
        "agent-automation",
        "research"
      ],
      "sourceRefs": [
        {
          "id": "official-site:gpt-5-mini-official",
          "label": "GPT-5 Mini official",
          "url": "https://openai.com/api/",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:gpt-5-mini-docs",
          "label": "GPT-5 Mini docs",
          "url": "https://platform.openai.com/docs/models",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "pricing-page:gpt-5-mini-pricing",
          "label": "GPT-5 Mini pricing",
          "url": "https://openai.com/api/pricing/",
          "kind": "pricing-page",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "pricing"
          ]
        },
        {
          "id": "manual-review:gpt-5-mini-verdictlens-review",
          "label": "GPT-5 Mini VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "claude-3-7-sonnet",
      "name": "Claude 3.7 Sonnet",
      "family": "Claude 3.7",
      "providerId": "anthropic",
      "officialUrl": "https://www.anthropic.com/claude",
      "docsUrl": "https://docs.anthropic.com/en/docs/about-claude/models/overview",
      "pricingUrl": "https://www.anthropic.com/pricing#api",
      "description": {
        "en": "Highly trusted reasoning and coding model with exceptional writing quality and calm, consistent outputs.",
        "zh-TW": "高信任度的推理與程式模型，寫作品質優秀，輸出穩定且一致性高。"
      },
      "modalities": [
        "text",
        "image",
        "tools"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "image",
          "file"
        ],
        "output": [
          "text",
          "json"
        ]
      },
      "contextWindow": "200K",
      "pricing": {
        "input": "$3 / 1M tok",
        "output": "$15 / 1M tok",
        "inputUsdPerMillion": "3",
        "outputUsdPerMillion": "15"
      },
      "speedCategory": "Balanced",
      "toolSupport": {
        "functionCalling": true,
        "structuredOutput": true,
        "remoteMcp": true
      },
      "overallScore": 91,
      "scores": {
        "capability": 94,
        "useCaseFitness": 92,
        "costEfficiency": 82,
        "speed": 84,
        "reliability": 95,
        "agentReadiness": 91,
        "ecosystem": 90
      },
      "bestUseCases": [
        "coding",
        "research"
      ],
      "worksWith": [
        "Claude Code",
        "LangGraph",
        "Playwright",
        "MCP servers"
      ],
      "tags": [
        "reasoning",
        "coding",
        "writing"
      ],
      "strengths": [
        {
          "en": "Excellent code edits and technical explanations",
          "zh-TW": "程式修改與技術解說能力出色"
        },
        {
          "en": "Strong output consistency for editorial workflows",
          "zh-TW": "在編輯與寫作流程中穩定度高"
        }
      ],
      "caveats": [
        {
          "en": "Tool ecosystem is strong but narrower than the broadest platform stacks",
          "zh-TW": "工具生態不錯，但仍略窄於最完整的平台型生態"
        },
        {
          "en": "Latency can rise on very long deliberative prompts",
          "zh-TW": "超長推理提示下延遲可能上升"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:claude-3-7-sonnet",
      "status": "active",
      "displayName": "Claude 3.7 Sonnet",
      "summary": {
        "en": "Highly trusted reasoning and coding model with exceptional writing quality and calm, consistent outputs.",
        "zh-TW": "高信任度的推理與程式模型，寫作品質優秀，輸出穩定且一致性高。"
      },
      "bestFor": [
        "coding",
        "research"
      ],
      "sourceRefs": [
        {
          "id": "official-site:claude-3-7-sonnet-official",
          "label": "Claude 3.7 Sonnet official",
          "url": "https://www.anthropic.com/claude",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:claude-3-7-sonnet-docs",
          "label": "Claude 3.7 Sonnet docs",
          "url": "https://docs.anthropic.com/en/docs/about-claude/models/overview",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "pricing-page:claude-3-7-sonnet-pricing",
          "label": "Claude 3.7 Sonnet pricing",
          "url": "https://www.anthropic.com/pricing#api",
          "kind": "pricing-page",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "pricing"
          ]
        },
        {
          "id": "manual-review:claude-3-7-sonnet-verdictlens-review",
          "label": "Claude 3.7 Sonnet VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "claude-3-5-haiku",
      "name": "Claude 3.5 Haiku",
      "family": "Claude 3.5",
      "providerId": "anthropic",
      "officialUrl": "https://www.anthropic.com/claude",
      "docsUrl": "https://docs.anthropic.com/en/docs/about-claude/models/overview",
      "pricingUrl": "https://www.anthropic.com/pricing#api",
      "description": {
        "en": "Fast and lightweight model ideal for summarization, routing, support tasks, and budget-sensitive assistants.",
        "zh-TW": "快速輕量模型，適合摘要、任務路由、客服與成本敏感型助理。"
      },
      "modalities": [
        "text",
        "image"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "image",
          "file"
        ],
        "output": [
          "text",
          "json"
        ]
      },
      "contextWindow": "200K",
      "pricing": {
        "input": "$0.8 / 1M tok",
        "output": "$4 / 1M tok",
        "inputUsdPerMillion": "0.8",
        "outputUsdPerMillion": "4"
      },
      "speedCategory": "Fast",
      "toolSupport": {
        "functionCalling": true,
        "structuredOutput": true
      },
      "overallScore": 84,
      "scores": {
        "capability": 78,
        "useCaseFitness": 85,
        "costEfficiency": 92,
        "speed": 96,
        "reliability": 88,
        "agentReadiness": 80,
        "ecosystem": 85
      },
      "bestUseCases": [
        "research",
        "agent-automation"
      ],
      "worksWith": [
        "LangGraph",
        "SerpApi",
        "Slack assistants"
      ],
      "tags": [
        "fast",
        "budget",
        "summaries"
      ],
      "strengths": [
        {
          "en": "Responsive and economical for high-volume operations",
          "zh-TW": "高量操作下反應快且成本低"
        },
        {
          "en": "Strong default tone quality for concise outputs",
          "zh-TW": "簡潔輸出時語氣表現穩定"
        }
      ],
      "caveats": [
        {
          "en": "Limited ceiling for complex multi-step planning",
          "zh-TW": "複雜多步驟規劃能力上限有限"
        },
        {
          "en": "Less appealing for agent workflows that require richer tool state",
          "zh-TW": "需要更複雜工具狀態的 agent 流程較不適合"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:claude-3-5-haiku",
      "status": "active",
      "displayName": "Claude 3.5 Haiku",
      "summary": {
        "en": "Fast and lightweight model ideal for summarization, routing, support tasks, and budget-sensitive assistants.",
        "zh-TW": "快速輕量模型，適合摘要、任務路由、客服與成本敏感型助理。"
      },
      "bestFor": [
        "research",
        "agent-automation"
      ],
      "sourceRefs": [
        {
          "id": "official-site:claude-3-5-haiku-official",
          "label": "Claude 3.5 Haiku official",
          "url": "https://www.anthropic.com/claude",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:claude-3-5-haiku-docs",
          "label": "Claude 3.5 Haiku docs",
          "url": "https://docs.anthropic.com/en/docs/about-claude/models/overview",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "pricing-page:claude-3-5-haiku-pricing",
          "label": "Claude 3.5 Haiku pricing",
          "url": "https://www.anthropic.com/pricing#api",
          "kind": "pricing-page",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "pricing"
          ]
        },
        {
          "id": "manual-review:claude-3-5-haiku-verdictlens-review",
          "label": "Claude 3.5 Haiku VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "gemini-2-5-pro",
      "name": "Gemini 2.5 Pro",
      "family": "Gemini 2.5",
      "providerId": "google",
      "officialUrl": "https://ai.google.dev/gemini-api",
      "docsUrl": "https://ai.google.dev/gemini-api/docs/models",
      "pricingUrl": "https://ai.google.dev/gemini-api/docs/pricing",
      "description": {
        "en": "Powerful multimodal model with strong long-context analysis, research workflows, and broad document understanding.",
        "zh-TW": "強大的多模態模型，擅長長上下文分析、研究流程與複雜文件理解。"
      },
      "modalities": [
        "text",
        "image",
        "audio",
        "video",
        "tools"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "image",
          "audio",
          "video",
          "file"
        ],
        "output": [
          "text",
          "json"
        ]
      },
      "contextWindow": "1M",
      "pricing": {
        "input": "$3.5 / 1M tok",
        "output": "$10.5 / 1M tok",
        "inputUsdPerMillion": "3.5",
        "outputUsdPerMillion": "10.5"
      },
      "speedCategory": "Balanced",
      "toolSupport": {
        "functionCalling": true,
        "structuredOutput": true,
        "webSearch": true,
        "codeExecution": true
      },
      "overallScore": 90,
      "scores": {
        "capability": 95,
        "useCaseFitness": 92,
        "costEfficiency": 84,
        "speed": 81,
        "reliability": 89,
        "agentReadiness": 88,
        "ecosystem": 92
      },
      "bestUseCases": [
        "research",
        "agent-automation"
      ],
      "worksWith": [
        "Notebook workflows",
        "Document analysis",
        "n8n",
        "Vertex AI stacks"
      ],
      "tags": [
        "long-context",
        "multimodal",
        "research"
      ],
      "strengths": [
        {
          "en": "Outstanding large-document synthesis",
          "zh-TW": "大型文件整合與歸納能力突出"
        },
        {
          "en": "One of the best multimodal coverage profiles",
          "zh-TW": "多模態覆蓋面非常完整"
        }
      ],
      "caveats": [
        {
          "en": "UX across endpoints can feel less uniform than the most polished competitors",
          "zh-TW": "不同端點體驗的一致性略遜於最成熟對手"
        },
        {
          "en": "Not always the fastest interactive pick",
          "zh-TW": "互動式回應未必是最快選擇"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:gemini-2-5-pro",
      "status": "active",
      "displayName": "Gemini 2.5 Pro",
      "summary": {
        "en": "Powerful multimodal model with strong long-context analysis, research workflows, and broad document understanding.",
        "zh-TW": "強大的多模態模型，擅長長上下文分析、研究流程與複雜文件理解。"
      },
      "bestFor": [
        "research",
        "agent-automation"
      ],
      "sourceRefs": [
        {
          "id": "official-site:gemini-2-5-pro-official",
          "label": "Gemini 2.5 Pro official",
          "url": "https://ai.google.dev/gemini-api",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:gemini-2-5-pro-docs",
          "label": "Gemini 2.5 Pro docs",
          "url": "https://ai.google.dev/gemini-api/docs/models",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "pricing-page:gemini-2-5-pro-pricing",
          "label": "Gemini 2.5 Pro pricing",
          "url": "https://ai.google.dev/gemini-api/docs/pricing",
          "kind": "pricing-page",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "pricing"
          ]
        },
        {
          "id": "manual-review:gemini-2-5-pro-verdictlens-review",
          "label": "Gemini 2.5 Pro VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "gemini-2-5-flash",
      "name": "Gemini 2.5 Flash",
      "family": "Gemini 2.5",
      "providerId": "google",
      "officialUrl": "https://ai.google.dev/gemini-api",
      "docsUrl": "https://ai.google.dev/gemini-api/docs/models",
      "pricingUrl": "https://ai.google.dev/gemini-api/docs/pricing",
      "description": {
        "en": "Low-latency multimodal workhorse for assistant surfaces, routing layers, and lightweight agent execution.",
        "zh-TW": "低延遲多模態工作馬，適合助理介面、路由層與輕量 agent 執行。"
      },
      "modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "image",
          "audio",
          "video",
          "file"
        ],
        "output": [
          "text",
          "json"
        ]
      },
      "contextWindow": "1M",
      "pricing": {
        "input": "$0.7 / 1M tok",
        "output": "$2.8 / 1M tok",
        "inputUsdPerMillion": "0.7",
        "outputUsdPerMillion": "2.8"
      },
      "speedCategory": "Fast",
      "toolSupport": {
        "functionCalling": true,
        "structuredOutput": true,
        "webSearch": true
      },
      "overallScore": 86,
      "scores": {
        "capability": 82,
        "useCaseFitness": 86,
        "costEfficiency": 95,
        "speed": 97,
        "reliability": 86,
        "agentReadiness": 84,
        "ecosystem": 90
      },
      "bestUseCases": [
        "agent-automation",
        "research"
      ],
      "worksWith": [
        "Routing layers",
        "Multimodal inboxes",
        "n8n",
        "Vertex AI stacks"
      ],
      "tags": [
        "multimodal",
        "fast",
        "assistant"
      ],
      "strengths": [
        {
          "en": "Excellent latency for multimodal interfaces",
          "zh-TW": "多模態介面的延遲表現很優秀"
        },
        {
          "en": "Very compelling value for large-scale ingestion",
          "zh-TW": "大規模資料吞吐時成本效益佳"
        }
      ],
      "caveats": [
        {
          "en": "Reasoning depth trails premium frontier models",
          "zh-TW": "推理深度仍落後旗艦級模型"
        },
        {
          "en": "May need a higher-tier fallback for sensitive outputs",
          "zh-TW": "高敏感輸出情境可能需要更高階備援模型"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:gemini-2-5-flash",
      "status": "active",
      "displayName": "Gemini 2.5 Flash",
      "summary": {
        "en": "Low-latency multimodal workhorse for assistant surfaces, routing layers, and lightweight agent execution.",
        "zh-TW": "低延遲多模態工作馬，適合助理介面、路由層與輕量 agent 執行。"
      },
      "bestFor": [
        "agent-automation",
        "research"
      ],
      "sourceRefs": [
        {
          "id": "official-site:gemini-2-5-flash-official",
          "label": "Gemini 2.5 Flash official",
          "url": "https://ai.google.dev/gemini-api",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:gemini-2-5-flash-docs",
          "label": "Gemini 2.5 Flash docs",
          "url": "https://ai.google.dev/gemini-api/docs/models",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "pricing-page:gemini-2-5-flash-pricing",
          "label": "Gemini 2.5 Flash pricing",
          "url": "https://ai.google.dev/gemini-api/docs/pricing",
          "kind": "pricing-page",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "pricing"
          ]
        },
        {
          "id": "manual-review:gemini-2-5-flash-verdictlens-review",
          "label": "Gemini 2.5 Flash VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "grok-3-beta",
      "name": "Grok 3 Beta",
      "family": "Grok 3",
      "providerId": "xai",
      "officialUrl": "https://x.ai/",
      "docsUrl": "https://docs.x.ai/",
      "description": {
        "en": "Strong real-time orientation and increasingly capable reasoning model with distinctive web freshness advantages.",
        "zh-TW": "偏向即時資訊場景、推理能力持續增強，且具備新鮮網路資訊優勢的模型。"
      },
      "modalities": [
        "text",
        "image",
        "tools"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "image",
          "file"
        ],
        "output": [
          "text"
        ]
      },
      "contextWindow": "256K",
      "pricing": {
        "input": "$5 / 1M tok",
        "output": "$15 / 1M tok",
        "inputUsdPerMillion": "5",
        "outputUsdPerMillion": "15"
      },
      "speedCategory": "Balanced",
      "toolSupport": {
        "functionCalling": true,
        "webSearch": true
      },
      "overallScore": 83,
      "scores": {
        "capability": 86,
        "useCaseFitness": 84,
        "costEfficiency": 74,
        "speed": 86,
        "reliability": 80,
        "agentReadiness": 78,
        "ecosystem": 79
      },
      "bestUseCases": [
        "research"
      ],
      "worksWith": [
        "Fresh web scans",
        "Market monitoring",
        "social-context workflows"
      ],
      "tags": [
        "web-aware",
        "analysis",
        "realtime"
      ],
      "strengths": [
        {
          "en": "Useful when freshness matters",
          "zh-TW": "重視資訊新鮮度時很有價值"
        },
        {
          "en": "Confident long-form synthesis style",
          "zh-TW": "長文整合輸出風格有辨識度"
        }
      ],
      "caveats": [
        {
          "en": "Ecosystem and enterprise controls still maturing",
          "zh-TW": "生態與企業級控制面仍在成熟中"
        },
        {
          "en": "Less proven for complex agent loops",
          "zh-TW": "複雜 agent loop 的實戰驗證較少"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "status": "preview",
      "id": "model:grok-3-beta",
      "displayName": "Grok 3 Beta",
      "summary": {
        "en": "Strong real-time orientation and increasingly capable reasoning model with distinctive web freshness advantages.",
        "zh-TW": "偏向即時資訊場景、推理能力持續增強，且具備新鮮網路資訊優勢的模型。"
      },
      "bestFor": [
        "research"
      ],
      "sourceRefs": [
        {
          "id": "official-site:grok-3-beta-official",
          "label": "Grok 3 Beta official",
          "url": "https://x.ai/",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:grok-3-beta-docs",
          "label": "Grok 3 Beta docs",
          "url": "https://docs.x.ai/",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "manual-review:grok-3-beta-verdictlens-review",
          "label": "Grok 3 Beta VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "deepseek-r1",
      "name": "DeepSeek R1",
      "family": "DeepSeek",
      "providerId": "deepseek",
      "officialUrl": "https://www.deepseek.com/",
      "docsUrl": "https://api-docs.deepseek.com/",
      "description": {
        "en": "High-value reasoning model that punches above its price tier for technical problem solving and analytical depth.",
        "zh-TW": "高性價比推理模型，在技術解題與分析深度上常有超出價格帶的表現。"
      },
      "modalities": [
        "text"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "file"
        ],
        "output": [
          "text"
        ]
      },
      "contextWindow": "128K",
      "pricing": {
        "input": "$0.55 / 1M tok",
        "output": "$2.2 / 1M tok",
        "inputUsdPerMillion": "0.55",
        "outputUsdPerMillion": "2.2"
      },
      "speedCategory": "Deliberate",
      "toolSupport": {
        "functionCalling": true
      },
      "overallScore": 85,
      "scores": {
        "capability": 90,
        "useCaseFitness": 86,
        "costEfficiency": 97,
        "speed": 67,
        "reliability": 81,
        "agentReadiness": 76,
        "ecosystem": 73
      },
      "bestUseCases": [
        "coding",
        "research"
      ],
      "worksWith": [
        "Cost-sensitive reasoning stacks",
        "batch analysis",
        "fallback reasoning lanes"
      ],
      "tags": [
        "reasoning",
        "value",
        "math"
      ],
      "strengths": [
        {
          "en": "Very attractive for reasoning-heavy workloads on a budget",
          "zh-TW": "預算有限時做重推理任務非常划算"
        },
        {
          "en": "Strong analytical style for math and code",
          "zh-TW": "在數學與程式分析上表現強勢"
        }
      ],
      "caveats": [
        {
          "en": "Latency and ecosystem maturity remain trade-offs",
          "zh-TW": "延遲與生態成熟度仍是代價"
        },
        {
          "en": "Operational guarantees vary by deployment path",
          "zh-TW": "不同部署路徑下的穩定性保證不一"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:deepseek-r1",
      "status": "active",
      "displayName": "DeepSeek R1",
      "summary": {
        "en": "High-value reasoning model that punches above its price tier for technical problem solving and analytical depth.",
        "zh-TW": "高性價比推理模型，在技術解題與分析深度上常有超出價格帶的表現。"
      },
      "bestFor": [
        "coding",
        "research"
      ],
      "sourceRefs": [
        {
          "id": "official-site:deepseek-r1-official",
          "label": "DeepSeek R1 official",
          "url": "https://www.deepseek.com/",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:deepseek-r1-docs",
          "label": "DeepSeek R1 docs",
          "url": "https://api-docs.deepseek.com/",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "manual-review:deepseek-r1-verdictlens-review",
          "label": "DeepSeek R1 VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "llama-4-maverick",
      "name": "Llama 4 Maverick",
      "family": "Llama 4",
      "providerId": "meta",
      "officialUrl": "https://ai.meta.com/llama/",
      "docsUrl": "https://ai.meta.com/llama/",
      "description": {
        "en": "Flexible open-weight option with broad community experimentation and strong customization potential.",
        "zh-TW": "彈性高的開放權重模型，社群實驗豐富，客製化潛力強。"
      },
      "modalities": [
        "text",
        "image"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "image",
          "file"
        ],
        "output": [
          "text"
        ]
      },
      "contextWindow": "256K",
      "pricing": {
        "input": "Self-host / variable",
        "output": "Self-host / variable",
        "notes": "Depends on infra and serving path."
      },
      "speedCategory": "Balanced",
      "toolSupport": {
        "functionCalling": true
      },
      "overallScore": 80,
      "scores": {
        "capability": 84,
        "useCaseFitness": 82,
        "costEfficiency": 88,
        "speed": 79,
        "reliability": 77,
        "agentReadiness": 75,
        "ecosystem": 89
      },
      "bestUseCases": [
        "agent-automation",
        "research"
      ],
      "worksWith": [
        "self-hosted inference",
        "vector retrieval",
        "custom fine-tuning"
      ],
      "tags": [
        "open-weight",
        "customizable",
        "self-hosted"
      ],
      "strengths": [
        {
          "en": "Strong fit for teams that need control and self-hosting",
          "zh-TW": "需要掌控權與自架的團隊會很適合"
        },
        {
          "en": "Community momentum creates many integration paths",
          "zh-TW": "社群動能帶來豐富整合方式"
        }
      ],
      "caveats": [
        {
          "en": "Production quality depends heavily on deployment discipline",
          "zh-TW": "正式環境品質非常依賴部署水準"
        },
        {
          "en": "Benchmark strength does not always translate to polished UX",
          "zh-TW": "benchmark 成績不一定等於實際體驗成熟"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:llama-4-maverick",
      "status": "active",
      "displayName": "Llama 4 Maverick",
      "summary": {
        "en": "Flexible open-weight option with broad community experimentation and strong customization potential.",
        "zh-TW": "彈性高的開放權重模型，社群實驗豐富，客製化潛力強。"
      },
      "bestFor": [
        "agent-automation",
        "research"
      ],
      "sourceRefs": [
        {
          "id": "official-site:llama-4-maverick-official",
          "label": "Llama 4 Maverick official",
          "url": "https://ai.meta.com/llama/",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:llama-4-maverick-docs",
          "label": "Llama 4 Maverick docs",
          "url": "https://ai.meta.com/llama/",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "manual-review:llama-4-maverick-verdictlens-review",
          "label": "Llama 4 Maverick VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "qwen-3-max",
      "name": "Qwen 3 Max",
      "family": "Qwen 3",
      "providerId": "alibaba-cloud",
      "officialUrl": "https://www.alibabacloud.com/product/model-studio",
      "docsUrl": "https://www.alibabacloud.com/help/en/model-studio/",
      "description": {
        "en": "Ambitious frontier contender with strong multilingual performance and good enterprise utility across Asian markets.",
        "zh-TW": "具競爭力的前沿模型，在多語言尤其亞洲市場企業場景中有不錯表現。"
      },
      "modalities": [
        "text",
        "image",
        "tools"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "image",
          "file"
        ],
        "output": [
          "text",
          "json"
        ]
      },
      "contextWindow": "256K",
      "pricing": {
        "input": "$2.4 / 1M tok",
        "output": "$9 / 1M tok",
        "inputUsdPerMillion": "2.4",
        "outputUsdPerMillion": "9"
      },
      "speedCategory": "Balanced",
      "toolSupport": {
        "functionCalling": true,
        "structuredOutput": true
      },
      "overallScore": 84,
      "scores": {
        "capability": 88,
        "useCaseFitness": 85,
        "costEfficiency": 86,
        "speed": 82,
        "reliability": 82,
        "agentReadiness": 80,
        "ecosystem": 81
      },
      "bestUseCases": [
        "research",
        "agent-automation"
      ],
      "worksWith": [
        "multilingual support workflows",
        "enterprise copilots",
        "Asian-market products"
      ],
      "tags": [
        "multilingual",
        "enterprise",
        "asia"
      ],
      "strengths": [
        {
          "en": "Strong multilingual coverage and pragmatic enterprise fit",
          "zh-TW": "多語能力強，企業落地實用度高"
        },
        {
          "en": "Balanced performance profile without extreme weaknesses",
          "zh-TW": "整體表現均衡，沒有明顯短板"
        }
      ],
      "caveats": [
        {
          "en": "Global ecosystem mindshare is still catching up",
          "zh-TW": "全球生態心智仍在追趕中"
        },
        {
          "en": "Docs and examples can vary by region",
          "zh-TW": "文件與案例完整度因地區而異"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:qwen-3-max",
      "status": "active",
      "displayName": "Qwen 3 Max",
      "summary": {
        "en": "Ambitious frontier contender with strong multilingual performance and good enterprise utility across Asian markets.",
        "zh-TW": "具競爭力的前沿模型，在多語言尤其亞洲市場企業場景中有不錯表現。"
      },
      "bestFor": [
        "research",
        "agent-automation"
      ],
      "sourceRefs": [
        {
          "id": "official-site:qwen-3-max-official",
          "label": "Qwen 3 Max official",
          "url": "https://www.alibabacloud.com/product/model-studio",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:qwen-3-max-docs",
          "label": "Qwen 3 Max docs",
          "url": "https://www.alibabacloud.com/help/en/model-studio/",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "manual-review:qwen-3-max-verdictlens-review",
          "label": "Qwen 3 Max VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "mistral-large-2",
      "name": "Mistral Large 2",
      "family": "Mistral Large",
      "providerId": "mistral",
      "officialUrl": "https://mistral.ai/products/la-plateforme",
      "docsUrl": "https://docs.mistral.ai/",
      "pricingUrl": "https://mistral.ai/pricing",
      "description": {
        "en": "European flagship model with solid reasoning, concise style, and attractive deployment flexibility.",
        "zh-TW": "歐洲旗艦模型，具備紮實推理、精簡風格與靈活部署優勢。"
      },
      "modalities": [
        "text",
        "image"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "image",
          "file"
        ],
        "output": [
          "text",
          "json"
        ]
      },
      "contextWindow": "128K",
      "pricing": {
        "input": "$2 / 1M tok",
        "output": "$6 / 1M tok",
        "inputUsdPerMillion": "2",
        "outputUsdPerMillion": "6"
      },
      "speedCategory": "Balanced",
      "toolSupport": {
        "functionCalling": true,
        "structuredOutput": true
      },
      "overallScore": 82,
      "scores": {
        "capability": 84,
        "useCaseFitness": 83,
        "costEfficiency": 87,
        "speed": 84,
        "reliability": 81,
        "agentReadiness": 78,
        "ecosystem": 78
      },
      "bestUseCases": [
        "research",
        "coding"
      ],
      "worksWith": [
        "EU deployment needs",
        "internal copilots",
        "API-first stacks"
      ],
      "tags": [
        "enterprise",
        "concise",
        "eu"
      ],
      "strengths": [
        {
          "en": "Clean enterprise-friendly posture and deployment options",
          "zh-TW": "企業友善姿態與部署選項清楚"
        },
        {
          "en": "Good balance between cost and reasoning quality",
          "zh-TW": "成本與推理品質之間取得不錯平衡"
        }
      ],
      "caveats": [
        {
          "en": "Ecosystem is smaller than the hyperscaler leaders",
          "zh-TW": "生態規模仍小於超大平台領先者"
        },
        {
          "en": "Less differentiated for multimodal-heavy product surfaces",
          "zh-TW": "對多模態產品面來說差異化較少"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:mistral-large-2",
      "status": "active",
      "displayName": "Mistral Large 2",
      "summary": {
        "en": "European flagship model with solid reasoning, concise style, and attractive deployment flexibility.",
        "zh-TW": "歐洲旗艦模型，具備紮實推理、精簡風格與靈活部署優勢。"
      },
      "bestFor": [
        "research",
        "coding"
      ],
      "sourceRefs": [
        {
          "id": "official-site:mistral-large-2-official",
          "label": "Mistral Large 2 official",
          "url": "https://mistral.ai/products/la-plateforme",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:mistral-large-2-docs",
          "label": "Mistral Large 2 docs",
          "url": "https://docs.mistral.ai/",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "pricing-page:mistral-large-2-pricing",
          "label": "Mistral Large 2 pricing",
          "url": "https://mistral.ai/pricing",
          "kind": "pricing-page",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "pricing"
          ]
        },
        {
          "id": "manual-review:mistral-large-2-verdictlens-review",
          "label": "Mistral Large 2 VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "command-a",
      "name": "Command A",
      "family": "Command",
      "providerId": "cohere",
      "officialUrl": "https://cohere.com/",
      "docsUrl": "https://docs.cohere.com/",
      "pricingUrl": "https://cohere.com/pricing",
      "description": {
        "en": "Enterprise-oriented model with strong retrieval posture, business language handling, and dependable workflow integration.",
        "zh-TW": "偏企業導向的模型，適合檢索增強、商務語言處理與穩定工作流整合。"
      },
      "modalities": [
        "text",
        "tools"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "file"
        ],
        "output": [
          "text",
          "json"
        ]
      },
      "contextWindow": "256K",
      "pricing": {
        "input": "$2 / 1M tok",
        "output": "$8 / 1M tok",
        "inputUsdPerMillion": "2",
        "outputUsdPerMillion": "8"
      },
      "speedCategory": "Balanced",
      "toolSupport": {
        "functionCalling": true,
        "structuredOutput": true
      },
      "overallScore": 81,
      "scores": {
        "capability": 82,
        "useCaseFitness": 84,
        "costEfficiency": 83,
        "speed": 82,
        "reliability": 84,
        "agentReadiness": 79,
        "ecosystem": 77
      },
      "bestUseCases": [
        "research",
        "agent-automation"
      ],
      "worksWith": [
        "RAG stacks",
        "enterprise search",
        "business writing workflows"
      ],
      "tags": [
        "enterprise",
        "retrieval",
        "workflow"
      ],
      "strengths": [
        {
          "en": "Strong fit for internal knowledge workflows",
          "zh-TW": "很適合內部知識工作流"
        },
        {
          "en": "Dependable enterprise integration story",
          "zh-TW": "企業整合敘事穩定可信"
        }
      ],
      "caveats": [
        {
          "en": "Less mindshare in cutting-edge coding circles",
          "zh-TW": "在最前沿 coding 圈中的心智較弱"
        },
        {
          "en": "Not the most exciting pick for multimodal experiences",
          "zh-TW": "多模態體驗不是它的主打強項"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:command-a",
      "status": "active",
      "displayName": "Command A",
      "summary": {
        "en": "Enterprise-oriented model with strong retrieval posture, business language handling, and dependable workflow integration.",
        "zh-TW": "偏企業導向的模型，適合檢索增強、商務語言處理與穩定工作流整合。"
      },
      "bestFor": [
        "research",
        "agent-automation"
      ],
      "sourceRefs": [
        {
          "id": "official-site:command-a-official",
          "label": "Command A official",
          "url": "https://cohere.com/",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:command-a-docs",
          "label": "Command A docs",
          "url": "https://docs.cohere.com/",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "pricing-page:command-a-pricing",
          "label": "Command A pricing",
          "url": "https://cohere.com/pricing",
          "kind": "pricing-page",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "pricing"
          ]
        },
        {
          "id": "manual-review:command-a-verdictlens-review",
          "label": "Command A VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    },
    {
      "slug": "perplexity-sonar-reasoning-pro",
      "name": "Sonar Reasoning Pro",
      "family": "Sonar",
      "providerId": "perplexity",
      "officialUrl": "https://www.perplexity.ai/hub/blog/sonar",
      "docsUrl": "https://docs.perplexity.ai/",
      "description": {
        "en": "Research-first model experience optimized for answer grounding, current web synthesis, and citation-friendly output.",
        "zh-TW": "偏研究導向的模型體驗，強調有根據的答案、即時網頁整合與引用友善輸出。"
      },
      "modalities": [
        "text",
        "web"
      ],
      "modalityProfile": {
        "input": [
          "text",
          "file"
        ],
        "output": [
          "text"
        ]
      },
      "contextWindow": "128K",
      "pricing": {
        "input": "$2 / 1M tok",
        "output": "$8 / 1M tok",
        "inputUsdPerMillion": "2",
        "outputUsdPerMillion": "8"
      },
      "speedCategory": "Balanced",
      "toolSupport": {
        "webSearch": true
      },
      "overallScore": 83,
      "scores": {
        "capability": 85,
        "useCaseFitness": 89,
        "costEfficiency": 82,
        "speed": 80,
        "reliability": 84,
        "agentReadiness": 74,
        "ecosystem": 76
      },
      "bestUseCases": [
        "research"
      ],
      "worksWith": [
        "citation-heavy briefs",
        "market scans",
        "answer-grounding workflows"
      ],
      "tags": [
        "research",
        "citations",
        "web"
      ],
      "strengths": [
        {
          "en": "Great for research UX with source visibility",
          "zh-TW": "研究型體驗下可見來源這點很有優勢"
        },
        {
          "en": "Useful for analyst workflows that need current context",
          "zh-TW": "需要即時背景的分析流程相當適合"
        }
      ],
      "caveats": [
        {
          "en": "Not a primary choice for tool-heavy coding agents",
          "zh-TW": "不適合做工具密集的 coding agent 主力"
        },
        {
          "en": "Less control than fully orchestrated platform APIs",
          "zh-TW": "相比完整平台 API，可控性較低"
        }
      ],
      "updatedAt": "2026-04-09T04:20:00+08:00",
      "id": "model:perplexity-sonar-reasoning-pro",
      "status": "active",
      "displayName": "Sonar Reasoning Pro",
      "summary": {
        "en": "Research-first model experience optimized for answer grounding, current web synthesis, and citation-friendly output.",
        "zh-TW": "偏研究導向的模型體驗，強調有根據的答案、即時網頁整合與引用友善輸出。"
      },
      "bestFor": [
        "research"
      ],
      "sourceRefs": [
        {
          "id": "official-site:sonar-reasoning-pro-official",
          "label": "Sonar Reasoning Pro official",
          "url": "https://www.perplexity.ai/hub/blog/sonar",
          "kind": "official-site",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "officialUrl"
          ]
        },
        {
          "id": "official-docs:sonar-reasoning-pro-docs",
          "label": "Sonar Reasoning Pro docs",
          "url": "https://docs.perplexity.ai/",
          "kind": "official-docs",
          "trustTier": 5,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "docsUrl"
          ]
        },
        {
          "id": "manual-review:sonar-reasoning-pro-verdictlens-review",
          "label": "Sonar Reasoning Pro VerdictLens review",
          "url": "https://verdictlens.local/manual-review",
          "kind": "manual-review",
          "trustTier": 3,
          "fetchedAt": "2026-04-09T04:20:00+08:00",
          "fieldPaths": [
            "bestFor",
            "worksWith",
            "strengths",
            "caveats"
          ]
        }
      ],
      "lastVerifiedAt": "2026-04-09T04:20:00+08:00"
    }
  ]
}