{
  "report_date": "2026-05-13",
  "title": "AI 日报 2026-05-13",
  "summary": "今日一手大厂发布较少，日报保留近一周内对工程团队更可执行的模型、API、Agent 工具和开源发布：语音实时模型、默认模型升级、金融 Agent 模板、Grok 连接器、模型迁移提醒，以及 Ollama、vLLM、Codex、LangChain 的近期版本信号。",
  "main_items": [
    {
      "title": "OpenAI 在 API 中发布新一代实时语音模型",
      "event_date": "2026-05-07",
      "url": "https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/",
      "source": "OpenAI",
      "tier": "T0",
      "entities": [
        "OpenAI",
        "Realtime API",
        "GPT-Realtime-2",
        "GPT-Realtime-Translate",
        "GPT-Realtime-Whisper"
      ],
      "summary": "OpenAI 发布 GPT-Realtime-2、GPT-Realtime-Translate 和 GPT-Realtime-Whisper，把实时语音从转写和回复扩展到可推理、可调用工具、可实时翻译的生产接口。",
      "bullets": [
        "GPT-Realtime-2 面向实时语音 Agent，支持更长上下文、并行工具调用提示、可调 reasoning effort 和更强恢复行为。",
        "GPT-Realtime-Translate 面向跨语言实时对话，GPT-Realtime-Whisper 面向低延迟流式转写；两者适合客服、会议、教育和现场活动等语音工作流。",
        "价格和可用性已在原文列出，团队可以直接评估语音输入/输出 token、按分钟翻译和转写成本。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "OpenAI 将 ChatGPT 默认模型升级为 GPT-5.5 Instant",
      "event_date": "2026-05-05",
      "url": "https://openai.com/index/gpt-5-5-instant/",
      "source": "OpenAI",
      "tier": "T0",
      "entities": [
        "OpenAI",
        "ChatGPT",
        "GPT-5.5 Instant",
        "chat-latest"
      ],
      "summary": "OpenAI 更新 ChatGPT 默认模型，并在 API 中以 chat-latest 提供 GPT-5.5 Instant，重点改进事实性、回答紧凑度、视觉/数学/科学能力和个性化上下文使用。",
      "bullets": [
        "原文称 GPT-5.5 Instant 正在替代 GPT-5.3 Instant 成为默认模型，并在 API 中通过 chat-latest 暴露。",
        "增强个性化会引用历史聊天、文件和已连接 Gmail 等上下文，同时引入 memory sources 让用户查看部分使用过的上下文来源。",
        "对产品团队而言，默认模型变化会影响用户对回答长度、个性化和检索触发时机的预期，灰度评估应覆盖已有提示词和客服/知识库场景。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "Anthropic 提高 Claude Code 与 Claude API 使用限额",
      "event_date": "2026-05-06",
      "url": "https://www.anthropic.com/news/higher-limits-spacex",
      "source": "Anthropic",
      "tier": "T0",
      "entities": [
        "Anthropic",
        "Claude Code",
        "Claude API",
        "SpaceX"
      ],
      "summary": "Anthropic 宣布与 SpaceX 的算力合作，并同步提高 Claude Code 与 Claude API 使用限额，直接影响重度编码 Agent 和 Opus API 用户的容量规划。",
      "bullets": [
        "Claude Code 的五小时限额面向 Pro、Max、Team 和 seat-based Enterprise 计划翻倍，并移除 Pro/Max 峰值时段限额下调。",
        "Claude Opus API rate limits 提高，适合此前受吞吐限制的批量代码审查、长任务 Agent 和企业内部工具。",
        "算力合作被明确用于提高 Claude Pro、Claude Max 和 API 容量，工程团队可以重新评估并发、队列和降级策略。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "Anthropic 发布金融服务 Agent 模板与 Microsoft 365 集成",
      "event_date": "2026-05-05",
      "url": "https://www.anthropic.com/news/finance-agents",
      "source": "Anthropic",
      "tier": "T0",
      "entities": [
        "Anthropic",
        "Claude Cowork",
        "Claude Code",
        "Claude Managed Agents",
        "MCP"
      ],
      "summary": "Anthropic 发布十个面向金融服务的 ready-to-run Agent 模板，并扩展 Excel、PowerPoint、Word、Outlook、连接器和 Moody's MCP app 等集成。",
      "bullets": [
        "模板覆盖 pitchbook、KYC、月结、估值复核、财报审阅、市场研究等工作，可作为 Claude Cowork/Claude Code 插件或 Managed Agents cookbook 使用。",
        "模板结构包含 skills、connectors 和 subagents，强调可按企业建模规范、风控政策和审批流程改造。",
        "这类行业模板把 Agent 交付从单点提示词推进到权限、凭据、审计和工具调用组合，值得内部平台团队拆解复用。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "xAI 在 Grok Web、iOS 和 Android 上推出 Connectors",
      "event_date": "2026-05-06",
      "url": "https://x.ai/news/grok-connectors",
      "source": "xAI",
      "tier": "T0",
      "entities": [
        "xAI",
        "Grok",
        "Connectors",
        "MCP"
      ],
      "summary": "xAI 发布 Grok Connectors，把 SharePoint、Outlook、OneDrive、Google Workspace、Notion、GitHub、Linear 和自定义 MCP 接入 Grok。",
      "bullets": [
        "连接器支持读取、搜索、综合和部分写入动作，让 Grok 直接处理邮件、文档、表格、日历、代码仓库和项目任务。",
        "GitHub 与 Linear 接入说明编码和项目管理场景正在进入通用助手入口，不再只停留在 IDE 或单独 bot。",
        "Bring Your Own MCP 让企业可接入内部知识库、私有 API 或内部 MCP 网关，需要同步评估权限隔离和审计。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "xAI 公布 2026-05-15 API 模型退休迁移清单",
      "event_date": "2026-05-07",
      "url": "https://docs.x.ai/developers/migration/may-15-retirement",
      "source": "xAI Docs",
      "tier": "T0",
      "entities": [
        "xAI",
        "Grok API",
        "Grok 4.3"
      ],
      "summary": "xAI 文档列出 2026-05-15 将退休的旧 Grok API 模型，并给出 Grok 4.3 和 grok-imagine-image-quality 等替代项。",
      "bullets": [
        "受影响模型包括 grok-4-1、grok-4、grok-code-fast-1、grok-3 和 grok-imagine-image-pro 等旧 API 名称。",
        "文档建议大多数文本、推理和代码工作负载迁移到 grok-4.3，并按 reasoning effort 调整延迟与成本。",
        "使用 xAI API 的服务应在 2026-05-15 12:00 PT 前完成模型名替换、回归测试和错误监控。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "Ollama v0.23.3 发布，继续修复 MLX 与更新流程",
      "event_date": "2026-05-12",
      "url": "https://github.com/ollama/ollama/releases/tag/v0.23.3",
      "source": "GitHub Releases",
      "tier": "T2",
      "entities": [
        "Ollama",
        "MLX",
        "local inference"
      ],
      "summary": "Ollama v0.23.3 是近期本地推理工具链的实际发布信号，集中在 MLX model push、imagegen runner、推理状态超时和 macOS 更新流程修复。",
      "bullets": [
        "发布说明列出 MLX 相关推送行为、线程亲和性和推理状态超时修复，影响 Apple Silicon 本地推理稳定性。",
        "同时包含 app 更新流程 hardening，适合依赖桌面/本地运行时分发的团队跟进验证。",
        "该条属于 GitHub Release 信号，不等同于新模型发布，但对本地模型运行链路有直接操作价值。"
      ],
      "image_urls": [],
      "importance": "notable",
      "editorial_category": "ai_industry"
    },
    {
      "title": "vLLM v0.20.2 修复 DeepSeek V4、gpt-oss 与 Qwen3-VL 路径",
      "event_date": "2026-05-10",
      "url": "https://github.com/vllm-project/vllm/releases/tag/v0.20.2",
      "source": "GitHub Releases",
      "tier": "T2",
      "entities": [
        "vLLM",
        "DeepSeek V4",
        "gpt-oss",
        "Qwen3-VL"
      ],
      "summary": "vLLM v0.20.2 是小型补丁版本，修复 DeepSeek V4 sparse attention 与 KV cache、gpt-oss MXFP4 torch.compile、Qwen3-VL deepstack 边界等问题。",
      "bullets": [
        "DeepSeek V4 相关修复覆盖 Hopper persistent topk、CUDA graph capture 和 V1 engine KV cache manager。",
        "gpt-oss MXFP4 在 torch.compile 下的 fake op 参数被补齐，Qwen3-VL 在高负载下的边界检查问题被移除。",
        "线上推理平台若运行这些模型族，应优先在灰度环境验证该补丁版本。"
      ],
      "image_urls": [],
      "importance": "notable",
      "editorial_category": "ai_industry"
    }
  ],
  "projects": [
    {
      "name": "OpenAI Codex",
      "description": "0.130.0 在 2026-05-08 发布，remote-control、插件元数据和 Windows sandbox 修复对自动化编码工作流有参考价值。",
      "url": "https://github.com/openai/codex/releases/tag/rust-v0.130.0",
      "domains": [],
      "importance": "general"
    },
    {
      "name": "LangChain",
      "description": "2026-05-07 相关发布回补反序列化和路径遍历安全修复；依赖旧版本线的 Agent/RAG 服务应安排升级。",
      "url": "https://github.com/langchain-ai/langchain/releases",
      "domains": [],
      "importance": "general"
    }
  ],
  "builder_observations": [],
  "schema_version": 1,
  "canonical_url": "https://jasonxzwen.github.io/ai-daily-cn/reports/2026/05/2026-05-13.html",
  "html_path": "reports/2026/05/2026-05-13.html",
  "generated_at": "2026-05-13T08:43:54.469Z",
  "hero_highlights": [],
  "github_trending": [],
  "stories": [
    {
      "story_id": "story-advancing-voice-intelligence-with-new-models-in-the-api",
      "title": "OpenAI 在 API 中发布新一代实时语音模型",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-05-07",
      "primary_entity": "OpenAI",
      "event_type": "launch",
      "object": "OpenAI 在 API 中发布新一代实时语音模型",
      "what_happened": "OpenAI 发布 GPT-Realtime-2、GPT-Realtime-Translate 和 GPT-Realtime-Whisper，把实时语音从转写和回复扩展到可推理、可调用工具、可实时翻译的生产接口。",
      "why_it_matters": "GPT-Realtime-Translate 面向跨语言实时对话，GPT-Realtime-Whisper 面向低延迟流式转写；两者适合客服、会议、教育和现场活动等语音工作流。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "OpenAI",
          "url": "https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "story-gpt-5-5-instant",
      "title": "OpenAI 将 ChatGPT 默认模型升级为 GPT-5.5 Instant",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-05-05",
      "primary_entity": "OpenAI",
      "event_type": "update",
      "object": "OpenAI 将 ChatGPT 默认模型升级为 GPT-5.5 Instant",
      "what_happened": "OpenAI 更新 ChatGPT 默认模型，并在 API 中以 chat-latest 提供 GPT-5.5 Instant，重点改进事实性、回答紧凑度、视觉/数学/科学能力和个性化上下文使用。",
      "why_it_matters": "增强个性化会引用历史聊天、文件和已连接 Gmail 等上下文，同时引入 memory sources 让用户查看部分使用过的上下文来源。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "OpenAI",
          "url": "https://openai.com/index/gpt-5-5-instant/",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "story-higher-limits-spacex",
      "title": "Anthropic 提高 Claude Code 与 Claude API 使用限额",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-05-06",
      "primary_entity": "Anthropic",
      "event_type": "update",
      "object": "Anthropic 提高 Claude Code 与 Claude API 使用限额",
      "what_happened": "Anthropic 宣布与 SpaceX 的算力合作，并同步提高 Claude Code 与 Claude API 使用限额，直接影响重度编码 Agent 和 Opus API 用户的容量规划。",
      "why_it_matters": "Claude Opus API rate limits 提高，适合此前受吞吐限制的批量代码审查、长任务 Agent 和企业内部工具。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "Anthropic",
          "url": "https://www.anthropic.com/news/higher-limits-spacex",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "story-finance-agents",
      "title": "Anthropic 发布金融服务 Agent 模板与 Microsoft 365 集成",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-05-05",
      "primary_entity": "Anthropic",
      "event_type": "launch",
      "object": "Anthropic 发布金融服务 Agent 模板与 Microsoft 365 集成",
      "what_happened": "Anthropic 发布十个面向金融服务的 ready-to-run Agent 模板，并扩展 Excel、PowerPoint、Word、Outlook、连接器和 Moody's MCP app 等集成。",
      "why_it_matters": "模板结构包含 skills、connectors 和 subagents，强调可按企业建模规范、风控政策和审批流程改造。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "Anthropic",
          "url": "https://www.anthropic.com/news/finance-agents",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "story-grok-connectors",
      "title": "xAI 在 Grok Web、iOS 和 Android 上推出 Connectors",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-05-06",
      "primary_entity": "xAI",
      "event_type": "launch",
      "object": "xAI 在 Grok Web、iOS 和 Android 上推出 Connectors",
      "what_happened": "xAI 发布 Grok Connectors，把 SharePoint、Outlook、OneDrive、Google Workspace、Notion、GitHub、Linear 和自定义 MCP 接入 Grok。",
      "why_it_matters": "GitHub 与 Linear 接入说明编码和项目管理场景正在进入通用助手入口，不再只停留在 IDE 或单独 bot。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "xAI",
          "url": "https://x.ai/news/grok-connectors",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "story-may-15-retirement",
      "title": "xAI 公布 2026-05-15 API 模型退休迁移清单",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-05-07",
      "primary_entity": "xAI Docs",
      "event_type": "update",
      "object": "xAI 公布 2026-05-15 API 模型退休迁移清单",
      "what_happened": "xAI 文档列出 2026-05-15 将退休的旧 Grok API 模型，并给出 Grok 4.3 和 grok-imagine-image-quality 等替代项。",
      "why_it_matters": "文档建议大多数文本、推理和代码工作负载迁移到 grok-4.3，并按 reasoning effort 调整延迟与成本。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "xAI Docs",
          "url": "https://docs.x.ai/developers/migration/may-15-retirement",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "story-ollama-ollama",
      "title": "Ollama v0.23.3 发布，继续修复 MLX 与更新流程",
      "importance": "notable",
      "trend": "AI industry",
      "event_date": "2026-05-12",
      "primary_entity": "GitHub Releases",
      "event_type": "launch",
      "object": "Ollama v0.23.3 发布，继续修复 MLX 与更新流程",
      "what_happened": "Ollama v0.23.3 是近期本地推理工具链的实际发布信号，集中在 MLX model push、imagegen runner、推理状态超时和 macOS 更新流程修复。",
      "why_it_matters": "同时包含 app 更新流程 hardening，适合依赖桌面/本地运行时分发的团队跟进验证。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "GitHub Releases",
          "url": "https://github.com/ollama/ollama/releases/tag/v0.23.3",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "story-vllm-project-vllm",
      "title": "vLLM v0.20.2 修复 DeepSeek V4、gpt-oss 与 Qwen3-VL 路径",
      "importance": "notable",
      "trend": "AI industry",
      "event_date": "2026-05-10",
      "primary_entity": "GitHub Releases",
      "event_type": "update",
      "object": "vLLM v0.20.2 修复 DeepSeek V4、gpt-oss 与 Qwen3-VL 路径",
      "what_happened": "vLLM v0.20.2 是小型补丁版本，修复 DeepSeek V4 sparse attention 与 KV cache、gpt-oss MXFP4 torch.compile、Qwen3-VL deepstack 边界等问题。",
      "why_it_matters": "gpt-oss MXFP4 在 torch.compile 下的 fake op 参数被补齐，Qwen3-VL 在高负载下的边界检查问题被移除。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "GitHub Releases",
          "url": "https://github.com/vllm-project/vllm/releases/tag/v0.20.2",
          "type": "source"
        }
      ]
    }
  ],
  "huggingface_trending": [],
  "model_releases": [],
  "hot_blogs": [],
  "chinese_media_dynamics": [],
  "daily_tracking": [],
  "official_org_updates": [],
  "evidence_assets": [],
  "quality_status": {
    "status": "ok",
    "public_note": "Core discovery checks completed without blocking degradation."
  }
}