{
  "generatedAt": "2026-07-08T02:10:00+08:00",
  "version": "3.0",
  "date": "2026-07-08",
  "url": "https://competitor-tracker.pages.dev/2026-07-08.html",
  "isLatest": true,
  "title": "竞品新能力周报 · 2026-07-08",
  "summary": "10个平台更新 · Novita 14款图像/视频模型7/22退役迁移Kling v3.0 · Fireworks GLM5.2/Kimi K2.7 Code/MiniMax M3/Qwen 3.7 Plus上线+Kimi K2.5/Qwen3.6Plus下线 · AWS SageMaker HyperPod DPD推理优化+HuggingFace一键集成 · Together AI VLM微调+Transformers v5 BYOM · RunPod Serverless 6种部署路径 · 阿里百炼老旧模型延期下线",
  "stats": {
    "totalPlatforms": 15,
    "updatedPlatforms": 10,
    "customizationItems": 20
  },
  "groups": [
    {
      "name": "国内平台",
      "platforms": [
        {
          "name": "硅基流动",
          "emoji": "🔷",
          "updated": false,
          "source": {
            "url": "https://docs.siliconflow.cn/cn/release-notes/overview",
            "fetchedAt": "2026-07-08T02:00:00+08:00"
          },
          "highlights": [],
          "customizationItems": []
        },
        {
          "name": "火山方舟",
          "emoji": "🌋",
          "updated": false,
          "source": {
            "url": "https://www.volcengine.com/docs/82379/1159177",
            "fetchedAt": "2026-07-08T02:05:00+08:00"
          },
          "highlights": [
            "SPA页面，web_fetch返回404，需browser_agent抓取，本期跳过（不阻塞流程）"
          ],
          "customizationItems": []
        },
        {
          "name": "火山机器学习平台",
          "emoji": "⛰️",
          "updated": false,
          "source": {
            "url": "https://www.volcengine.com/docs/6459/2206987",
            "fetchedAt": "2026-07-08T02:05:00+08:00"
          },
          "highlights": [
            "SPA页面，web_fetch返回404，需browser_agent抓取，本期跳过（不阻塞流程）"
          ],
          "customizationItems": []
        },
        {
          "name": "百度千帆",
          "emoji": "🔍",
          "updated": false,
          "source": {
            "url": "https://cloud.baidu.com/doc/qianfan/s/Mmh8l4qwj",
            "fetchedAt": "2026-07-08T02:06:00+08:00"
          },
          "highlights": [
            "SPA页面，放最后采集，需browser_agent抓取，本期跳过（不阻塞流程）"
          ],
          "customizationItems": []
        },
        {
          "name": "阿里百炼",
          "emoji": "🌐",
          "updated": true,
          "source": {
            "url": "https://help.aliyun.com/zh/model-studio/model-release-notes",
            "fetchedAt": "2026-07-08T02:01:00+08:00"
          },
          "highlights": [
            "7月6日：部分老旧模型延期下线通知，原计划下线的部分老旧模型延期处理",
            "6月30日：团队版新增共享用量包（跨坐席共享Credits弹性用量包）",
            "6月29日：Coding Plan Pro套餐新客首月特惠¥39.90 / 智能体托管运行时API上线（平台托管会话与工具执行）/ API Key加密存储与业务空间专属推理API域名升级",
            "6月28日：qwen-turbo资源包启动退市通知",
            "6月24日：新增Codex客户端接入（终端AI编程助手接入百炼）",
            "6月23日：知识检索服务上线（多知识库联合检索与混合排序）/ 知识问答服务上线（基于大模型结合检索生成回答）"
          ],
          "customizationItems": [
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "部分老旧模型延期下线通知",
              "desc": "7月6日发布公告，原计划下线的部分老旧模型延期处理，为用户提供更多迁移时间。建议关注官方公告了解具体涉及的模型列表和新下线时间",
              "url": "https://www.aliyun.com/notice/118420",
              "tags": ["模型下线", "延期", "老旧模型"]
            }
          ]
        },
        {
          "name": "PPIO",
          "emoji": "🟢",
          "updated": true,
          "source": {
            "url": "https://blog.ppio.com/",
            "fetchedAt": "2026-07-08T02:02:00+08:00"
          },
          "highlights": [
            "6月：首发上线智谱GLM-5.2开源旗舰模型，Coding能力开源SOTA，LMArena代码榜开源第一",
            "6月：首发上线Kimi K2.7 Code，Kimi迄今最智能的Coding模型，长程编码能力突破，平均token消耗减少30%",
            "6月：上线MiniMax M3，最高支持1M超长上下文，原生多模态，SWE-Bench Pro超过GPT-5.5和Gemini 3.1 Pro",
            "6月22日：PPIO入选36氪2026最具价值成长企业100榜单"
          ],
          "customizationItems": [
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "首发上线GLM-5.2",
              "desc": "首发上线智谱最新开源旗舰模型GLM-5.2。Coding能力开源SOTA，发布即获LMArena代码榜开源模型第一、全球模型第二，整体表现仅次于Claude Fable 5；支持真正可用的1M上下文，一次任务完成从需求到多端可部署产物的完整开发链路；构建逾万个可验证任务环境，覆盖九大主流编程语言",
              "url": "https://blog.ppio.com/ppioshou-fa-shang-xian-glm-5-2-dai-ma-neng-li-jin-ci-yu-claude-fable-5/",
              "tags": ["GLM-5.2", "Coding", "1M上下文", "开源SOTA"]
            },
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "首发上线Kimi K2.7 Code",
              "desc": "PPIO已上线Kimi K2.7 Code，这是Kimi迄今最智能的Coding模型。以编码为中心的智能体模型，在长上下文中更可靠地遵循指令，支持文本、图片与视频输入，思考模式，对话与Agent任务。相比K2.6大幅改善长程任务中的过度思考倾向，平均token消耗减少30%",
              "url": "https://blog.ppio.com/ppioshou-fa-shang-xian-kimi-k2-7-code-kimizui-zhi-neng-de-codingmo-xing/",
              "tags": ["Kimi-K2.7-Code", "Coding", "多模态", "token优化"]
            },
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "上线MiniMax M3",
              "desc": "作为MiniMax官方MaaS合作伙伴，PPIO首发上线MiniMax M3模型。在编程和智能体等专业任务上达到前沿能力，最高支持1M超长上下文，原生多模态模型支持图片和视频输入并能操作电脑桌面。SWE-Bench Pro超过GPT-5.5和Gemini 3.1 Pro，接近Opus 4.7",
              "url": "https://blog.ppio.com/ppioshang-xian-minimax-m3-xian-shi-7tian-5zhe-you-hui/",
              "tags": ["MiniMax-M3", "1M上下文", "多模态", "SWE-Bench"]
            }
          ]
        },
        {
          "name": "无问芯穹",
          "emoji": "♾️",
          "updated": true,
          "source": {
            "url": "https://docs.infini-ai.com/gen-studio/changelog.html",
            "fetchedAt": "2026-07-08T02:03:00+08:00"
          },
          "highlights": [
            "6月30日：14款模型下架，包括Qwen3全系列(8B/14B/30B-A3B/32B等)、deepseek-r1-distill-qwen-32b、deepseek-v3.2-exp、qwen3-vl-235b系列、qwen3-next-80b系列、qwen3-coder-480b、kimi-k2-thinking等",
            "迁移建议：Qwen3系列→qwen3.6，deepseek-r1-distill-qwen-32b/deepseek-v3.2-exp→deepseek-v4-flash或deepseek-v4-pro，kimi-k2-thinking→kimi-k2.6",
            "6月18日：上线Qwen3.6系列模型(qwen3.6-27b, qwen3.6-35b-a3b)",
            "6月17日：上线GLM-5.2",
            "6月16日：GenStudio API新增错误码10024(预算超限)，预置服务按量在线调用/视频大模型调用/批量推理支持预算管理"
          ],
          "customizationItems": [
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "14款模型下架及迁移指引",
              "desc": "6月30日下架14款模型：deepseek-r1-distill-qwen-32b、deepseek-v3.2-exp、qwen3-vl-235b-a22b-thinking/instruct、qwen3-next-80b-a3b-thinking/instruct、qwen3-coder-480b-a35b-instruct、qwen3-235b-a22b-instruct-2507/235b-a22b、qwen3-8b/14b/32b/30b-a3b、kimi-k2-thinking。迁移建议：Qwen3系列→qwen3.6，deepseek-r1-distill/deepseek-v3.2-exp→deepseek-v4-flash/v4-pro，kimi-k2-thinking→kimi-k2.6",
              "url": "https://docs.infini-ai.com/gen-studio/changelog.html",
              "tags": ["模型下架", "Qwen3", "DeepSeek", "Kimi-K2-Thinking", "迁移指引"]
            }
          ]
        }
      ]
    },
    {
      "name": "海外平台",
      "platforms": [
        {
          "name": "Fireworks",
          "emoji": "🎆",
          "updated": true,
          "source": {
            "url": "https://docs.fireworks.ai/updates/changelog",
            "fetchedAt": "2026-07-08T02:04:00+08:00"
          },
          "highlights": [
            "Serverless弃用：Kimi K2.5和Qwen 3.6 Plus，建议迁移到Kimi K2.6和Qwen 3.7 Plus",
            "Serverless弃用：MiniMax M2.5，建议迁移到MiniMax M2.7",
            "新模型上线：GLM 5.2 现已在模型库可用",
            "新模型上线：Kimi K2.7 Code、MiniMax M3、Qwen 3.7 Plus 现已在模型库可用",
            "音频推理和图像生成功能弃用"
          ],
          "customizationItems": [
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "Kimi K2.5和Qwen 3.6 Plus从Serverless弃用",
              "desc": "Kimi K2.5和Qwen 3.6 Plus从serverless弃用。迁移建议：Kimi K2.5→Kimi K2.6，Qwen 3.6 Plus→Qwen 3.7 Plus",
              "url": "https://docs.fireworks.ai/updates/changelog",
              "tags": ["模型下线", "Kimi-K2.5", "Qwen-3.6-Plus", "Serverless"]
            },
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "GLM 5.2上线模型库",
              "desc": "GLM 5.2现已在Fireworks模型库可用，支持API调用",
              "url": "https://app.fireworks.ai/models/fireworks/glm-5p2",
              "tags": ["GLM-5.2", "新模型", "模型库"]
            },
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "Kimi K2.7 Code、MiniMax M3、Qwen 3.7 Plus上线",
              "desc": "三款新模型现已在模型库可用：Kimi K2.7 Code（Kimi最智能的Coding模型）、MiniMax M3（1M上下文多模态）、Qwen 3.7 Plus",
              "url": "https://docs.fireworks.ai/updates/changelog",
              "tags": ["Kimi-K2.7-Code", "MiniMax-M3", "Qwen-3.7-Plus", "新模型"]
            }
          ]
        },
        {
          "name": "AWS Bedrock",
          "emoji": "📦",
          "updated": true,
          "source": {
            "url": "https://aws.amazon.com/about-aws/whats-new/recent/feed/",
            "fetchedAt": "2026-07-08T02:07:00+08:00"
          },
          "highlights": [
            "7月1日：Amazon Bedrock AgentCore提升默认运行时配额限制，US East/West支持5000并发会话，其他区域2500，200 interactions/second，25 new sessions/second",
            "7月1日：Amazon Bedrock AgentCore在4个新AWS区域可用（Asia Pacific Bangkok/Malaysia, Europe Milan/Spain）"
          ],
          "customizationItems": [
            {
              "category": "deploy",
              "categoryLabel": "部署",
              "title": "AgentCore提升默认运行时配额",
              "desc": "Amazon Bedrock AgentCore提升默认运行时配额限制，US East (N. Virginia)和US West (Oregon)支持5000个活跃并发会话，其他支持区域2500个。所有区域支持200 agent interactions/second和25 new sessions/second，客户可同时运行更多AI Agent并处理高吞吐工作负载",
              "url": "https://aws.amazon.com/about-aws/whats-new/2026/07/amazon-bedrock-agentcore-increases-default-runtime-quota-limits/",
              "tags": ["AgentCore", "配额提升", "并发会话", "高吞吐"]
            },
            {
              "category": "deploy",
              "categoryLabel": "部署",
              "title": "AgentCore在4个新区域可用",
              "desc": "Amazon Bedrock AgentCore在4个新AWS区域可用：Asia Pacific (Bangkok)、Asia Pacific (Malaysia)、Europe (Milan)、Europe (Spain)。客户可在这些区域更接近终端用户构建和运行Agent，降低延迟。包含agent runtime、身份访问控制、策略管理、会话持久化、工具连接和可观测性能力",
              "url": "https://aws.amazon.com/about-aws/whats-new/2026/06/amazon-bedrock-agentcore-four-additional-regions/",
              "tags": ["AgentCore", "新区域", "低延迟", "Agent运行时"]
            }
          ]
        },
        {
          "name": "AWS SageMaker AI",
          "emoji": "🧠",
          "updated": true,
          "source": {
            "url": "https://aws.amazon.com/about-aws/whats-new/recent/feed/",
            "fetchedAt": "2026-07-08T02:07:00+08:00"
          },
          "highlights": [
            "7月6日：Amazon SageMaker HyperPod支持分离式Prefill和Decode(DPD)推理优化，将prefill和decode分离到专用GPU池，KV cache通过EFA+GPU-Direct RDMA传输",
            "7月6日：Amazon SageMaker Studio集成Hugging Face，支持一键模型部署和定制（从模型发现到Studio环境一键直达）",
            "7月2日：Amazon SageMaker Unified Studio支持Terraform基础设施即代码部署",
            "7月2日：Amazon SageMaker HyperPod支持AMI版本管理和自动安全补丁（opt-in per-instance-group，仅应用向后兼容的安全补丁）"
          ],
          "customizationItems": [
            {
              "category": "deploy",
              "categoryLabel": "部署",
              "title": "HyperPod分离式Prefill和Decode(DPD)推理优化",
              "desc": "Amazon SageMaker HyperPod支持Disaggregated Prefill and Decode (DPD)，将LLM推理的prefill和decode两阶段分离到专用GPU池，KV cache通过EFA使用GPU-Direct RDMA传输。提供更一致的per-token延迟、更高吞吐，可独立扩展prefill和decode容量。智能路由器自动将长上下文请求导向分离路径，短prompt直接送decoder",
              "url": "https://aws.amazon.com/about-aws/whats-new/2026/7/amazon-sagemaker-hyperpod-dpd/",
              "tags": ["HyperPod", "DPD", "推理优化", "KV cache", "EFA", "GPU-Direct RDMA"]
            },
            {
              "category": "deploy",
              "categoryLabel": "部署",
              "title": "SageMaker Studio集成Hugging Face一键部署",
              "desc": "Amazon SageMaker Studio支持从Hugging Face直接集成，在Hugging Face任意支持模型上选择Customize on SageMaker AI或Deploy on SageMaker AI，一键直达对应工作流页面，模型预加载即用。新客户完成AWS注册即可获得预配置权限的Studio环境，支持serverless模型定制作业（含强化学习微调奖励函数）、模型评测和部署到SageMaker或Bedrock端点",
              "url": "https://aws.amazon.com/about-aws/whats-new/2026/07/sagemaker-studio-hugging-face-integration/",
              "tags": ["HuggingFace", "一键部署", "模型定制", "Studio集成"]
            },
            {
              "category": "deploy",
              "categoryLabel": "部署",
              "title": "Unified Studio支持Terraform+HyperPod AMI版本管理",
              "desc": "Amazon SageMaker Unified Studio支持Terraform部署，通过terraform-aws-sagemaker-unified-studio模块用版本控制模板部署domain。同时HyperPod支持AMI版本管理（语义化版本号、版本漂移检测、回滚）和自动安全补丁（opt-in per-instance-group，仅应用向后兼容补丁，不升级NVIDIA driver/CUDA等关键包的主次版本）",
              "url": "https://aws.amazon.com/about-aws/whats-new/2026/07/amazon-sagemaker-hyperpod-ami-version-auto-patch",
              "tags": ["Terraform", "AMI版本管理", "自动补丁", "基础设施即代码"]
            }
          ]
        },
        {
          "name": "Google Vertex AI",
          "emoji": "☁️",
          "updated": true,
          "source": {
            "url": "https://cloud.google.com/vertex-ai/docs/release-notes",
            "fetchedAt": "2026-07-08T02:08:00+08:00"
          },
          "highlights": [
            "6月30日：Agent Platform Workbench v2支持Python 3.12基础容器构建自定义容器（除默认Python 3.10外）",
            "5月26日：Colab Enterprise Data Science Agent正式可用，自动化探索性数据分析",
            "5月26日：Vertex AI Extensions弃用，将在2026年11月26日后关停，建议迁移到Agent Platform"
          ],
          "customizationItems": [
            {
              "category": "deploy",
              "categoryLabel": "部署",
              "title": "Agent Platform Workbench支持Python 3.12基础容器",
              "desc": "Agent Platform Workbench自定义容器支持Python 3.12基础容器（除默认Python 3.10外）。标准和slim基础容器可用：us-docker.pkg.dev/workbench-images/gcr.io/workbench-container-2606:latest 和 workbench-container-slim-2606:latest",
              "url": "https://cloud.google.com/vertex-ai/docs/release-notes",
              "tags": ["Python 3.12", "Workbench", "自定义容器", "基础镜像"]
            }
          ]
        },
        {
          "name": "Together AI",
          "emoji": "🤝",
          "updated": true,
          "source": {
            "url": "https://docs.together.ai/docs/changelog",
            "fetchedAt": "2026-07-08T02:09:00+08:00"
          },
          "highlights": [
            "新增微调模型：google/gemma-4-31B-it-VLM视觉语言模型支持微调",
            "BYOM微调支持Hugging Face Transformers v5或更早版本构建的模型",
            "新serverless模型：google/flash-image-3.1-lite(Gemini 3.1 Flash-Lite Image)、Qwen/Qwen3.6-35B-A3B-Lora(262K上下文)、alibaba/happyhorse-1.1-i2v/r2v/t2v(图生视频/参考生视频/文生视频)",
            "6月29日：Qwen/Qwen3.5-397B-A17B从serverless弃用",
            "6月25日：Seedance 2.0新增4K视频分辨率支持(3840x2160)"
          ],
          "customizationItems": [
            {
              "category": "fineTuning",
              "categoryLabel": "微调",
              "title": "VLM视觉语言模型微调+Transformers v5 BYOM",
              "desc": "新增google/gemma-4-31B-it-VLM视觉语言模型微调支持。同时BYOM(Bring Your Own Model)微调现支持Hugging Face Transformers v5或更早版本构建的模型，扩展自定义模型微调能力",
              "url": "https://docs.together.ai/docs/fine-tuning/supported-models",
              "tags": ["VLM微调", "gemma-4-31B", "视觉语言模型", "BYOM", "Transformers v5"]
            },
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "happyhorse 1.1视频模型+flash-image-3.1-lite上线",
              "desc": "新serverless模型：alibaba/happyhorse-1.1-i2v(图生视频)、happyhorse-1.1-r2v(参考生视频)、happyhorse-1.1-t2v(文生视频)；google/flash-image-3.1-lite(Gemini 3.1 Flash-Lite Image)；Qwen/Qwen3.6-35B-A3B-Lora(262,144上下文长度)",
              "url": "https://docs.together.ai/docs/serverless/models",
              "tags": ["happyhorse", "视频生成", "flash-image", "Gemini 3.1", "Qwen3.6-Lora"]
            },
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "Qwen3.5-397B-A17B从serverless弃用",
              "desc": "6月29日：Qwen/Qwen3.5-397B-A17B从serverless弃用，不再可用。推荐替代：MiniMaxAI/MiniMax-M3（作为on-demand dedicated endpoint可用）",
              "url": "https://docs.together.ai/docs/changelog",
              "tags": ["模型下线", "Qwen3.5-397B", "serverless"]
            }
          ]
        },
        {
          "name": "RunPod",
          "emoji": "🚀",
          "updated": true,
          "source": {
            "url": "https://docs.runpod.io/release-notes",
            "fetchedAt": "2026-07-08T02:10:00+08:00"
          },
          "highlights": [
            "7月1日：更新Serverless endpoint创建流程，提供6种部署路径（Hello World、Hugging Face LLM、Docker、GitHub、Flash、Hub）",
            "7月1日：新教程-用私有AWS ECR镜像部署Pods（BETA），通过cross-account IAM delegation拉取私有ECR仓库容器镜像"
          ],
          "customizationItems": [
            {
              "category": "deploy",
              "categoryLabel": "部署",
              "title": "Serverless endpoint 6种部署路径",
              "desc": "endpoint创建流程更新，提供6种部署路径：Hello World、Hugging Face LLM、Docker、GitHub、Flash、Hub（替代之前的Ready-to-Deploy Repos选项）。每种路径针对不同用例引导正确的设置方式",
              "url": "https://docs.runpod.io/serverless/endpoints/overview",
              "tags": ["Serverless", "部署路径", "HuggingFace", "Docker", "GitHub"]
            },
            {
              "category": "deploy",
              "categoryLabel": "部署",
              "title": "用私有AWS ECR镜像部署Pods (BETA)",
              "desc": "新教程覆盖如何通过cross-account IAM delegation从私有AWS ECR仓库拉取容器镜像到RunPod Pods。包含配置ECR仓库策略、在RunPod控制台添加ECR凭证、用私有镜像部署Pod，无需直接管理凭证",
              "url": "https://docs.runpod.io/tutorials/pods/use-private-ecr-images",
              "tags": ["AWS ECR", "私有镜像", "IAM", "BETA", "Pod部署"]
            }
          ]
        },
        {
          "name": "Novita AI",
          "emoji": "⚙️",
          "updated": true,
          "source": {
            "url": "https://novita.ai/docs/changelog/",
            "fetchedAt": "2026-07-08T02:11:00+08:00"
          },
          "highlights": [
            "7月7日：多款图像和视频模型将于7月22日从Novita平台退役",
            "受影响图像模型：Upscale、Remove Background、Replace Background、Inpainting、Remove Text、Cleanup、Image to Prompt（无替代方案）",
            "受影响视频模型：Kling V2.5 Turbo、Kling-o1、Kling V2.6 Pro、PixVerse V4.5、VIDU Q2/Q2 Turbo/Q2 Pro、Vidu Q3 Turbo/Q3 Pro → 迁移至Kling v3.0"
          ],
          "customizationItems": [
            {
              "category": "model",
              "categoryLabel": "模型",
              "title": "多款图像/视频模型7月22日退役",
              "desc": "7月7日发布公告，多款图像和视频模型将于7月22日从Novita平台退役。图像模型（Upscale/Remove Background/Replace Background/Inpainting/Remove Text/Cleanup/Image to Prompt）无替代方案；视频模型（Kling V2.5 Turbo/Kling-o1/Kling V2.6 Pro/PixVerse V4.5/VIDU Q2系列/Vidu Q3系列）迁移至Kling v3.0",
              "url": "https://novita.ai/docs/changelog/07-07-26",
              "tags": ["模型下线", "图像模型", "视频模型", "Kling v3.0", "PixVerse", "VIDU"]
            }
          ]
        },
        {
          "name": "OpenRouter",
          "emoji": "🔀",
          "updated": false,
          "source": {
            "url": "https://openrouter.ai/blog/",
            "fetchedAt": "2026-07-08T02:12:00+08:00"
          },
          "highlights": [],
          "customizationItems": []
        }
      ]
    }
  ]
}