Imagen 4 Fast 计划停服
Google 官方文档将 Imagen 4 标记为 Deprecated，并公布 2026-08-17 计划停服日期。
来源：Generate images using Imagen ↗把公司、产品、模型和版本放在同一条公开时间线上阅读。用发布、价格、可用性和生命周期变化，判断你的接入是否需要重新核验。
时间线记录帮助你回看公开变化及其证据，不替代模型推荐、采购结论或性能判断。
日期未公开的记录会保留在列表末尾，不根据上下文推断日期。
Google 官方文档将 Imagen 4 标记为 Deprecated，并公布 2026-08-17 计划停服日期。
来源：Generate images using Imagen ↗Google 官方文档将 Imagen 4 标记为 Deprecated，并公布 2026-08-17 计划停服日期。
来源：Generate images using Imagen ↗Google 官方文档将 Imagen 4 标记为 Deprecated，并公布 2026-08-17 计划停服日期。
来源：Generate images using Imagen ↗官方页面给出的 tentative retirement date 为 2026-08-05；这是 Anthropic API 日期，合作云平台可能有不同计划。
来源：Model deprecations | Claude API ↗该事件只记录来源发布行为。 AppFolio identifies Claude 3 Haiku as the high-volume, rapid-response model in Realm-X Messages. The vendor case reports almost tripled adoption after the Haiku and infrastructure changes, about 11 hours saved per property manager each week, more than half a million AI-suggested responses processed, and an average 26-second reduction in message response time.
来源：AppFolio and Claude Platform ↗该事件只记录来源发布行为。 The public case identifies Claude 3.7 Sonnet in Benchling's domain-specific evaluation and scientific assistants. It reports scientific question answering moving from hours to seconds, hours saved per study through data-entry automation, reduced transcription errors, and more than 100 organizations using the review feature.
来源：Benchling and Claude in Amazon Bedrock ↗该事件只记录来源发布行为。 The public case identifies Claude Opus 4.7 in Bolt's live design system agent. It reports an average 53-minute design system generation workflow, on-brand prototypes in roughly five minutes each, more than 10,000 users uploading design systems, and roughly 90% Agent SDK cache efficiency.
来源：Bolt and Claude Agent SDK ↗DeepSeek 官方 API 文档记录 v4-flash 已更新为 DeepSeek-V4-Flash-0731，调用方式保持不变；官网 2026-07-31 发布快照提供同日正式版 API 公测公告。
Anthropic 官方 API 生命周期页记录 Claude Opus 4 于 2026-06-15 退休，推荐迁移到 Claude Opus 4.8。
来源：Model deprecations | Claude API ↗Anthropic 官方 API 生命周期页记录 Claude Sonnet 4 于 2026-06-15 退休，推荐迁移到 Claude Sonnet 4.6。
来源：Model deprecations | Claude API ↗Mistral announced Mistral Medium 3.5 in public preview, remote coding agents in Vibe and a new Work mode in Le Chat.
来源：Remote agents in Vibe. Powered by Mistral Medium 3.5. ↗Mistral announced Small 4 as a unified model combining instruct, reasoning, multimodal and agentic coding capabilities.
来源：Introducing Mistral Small 4 ↗Anthropic 官方 API 生命周期页记录 Claude Haiku 3.5 于 2026-02-19 退休，推荐迁移到 Claude Haiku 4.5。
来源：Model deprecations | Claude API ↗Anthropic 官方 API 生命周期页记录 Claude Sonnet 3.7 于 2026-02-19 退休，推荐迁移到 Claude Sonnet 4.6。
来源：Model deprecations | Claude API ↗Mistral announced Devstral 2 and the Mistral Vibe CLI for code-agent workflows.
来源：Introducing: Devstral 2 and Mistral Vibe CLI. ↗Google lists December 2, 2025 as the earliest possible shutdown date and recommends gemini-3.1-pro-preview.
来源：Gemini deprecations ↗Anthropic 生命周期文档记录 Claude Sonnet 3.5 模型于 2025-10-28 退休，并推荐迁移到 Claude Sonnet 4.6。
来源：Model deprecations | Claude API ↗Z.ai 官方文章记录 GLM-4.5 于 2025-07-28 发布。
来源：GLM-4.5: Reasoning, Coding, and Agentic Abililties ↗腾讯混元官方仓库记录 Hunyuan-A13B-Instruct 及多个量化变体于 2025-06-27 开源。
来源：Tencent-Hunyuan/Hunyuan-A13B ↗Google's deprecations table lists June 17, 2025 as the release date for gemini-2.5-pro.
来源：Gemini deprecations ↗Qwen 官方发布 Qwen3，并在 Qwen Models 产品目录中公开多个 Dense 与 MoE 模型及其开放权重信息。
来源：Qwen3: Think Deeper, Act Faster ↗Qwen 官方发布 Qwen3，并将 Qwen3-235B-A22B 作为旗舰 MoE 模型。
来源：Qwen3: Think Deeper, Act Faster ↗DeepSeek 官方公告记录 DeepSeek-R1 及基于 Qwen2.5、Llama3 的六个蒸馏模型于 2025-01-20 开源。
来源：DeepSeek-R1 Release ↗Google Cloud lists gemini-1.5-pro-002 release date as September 24, 2024 in the Gemini Enterprise Agent Platform lifecycle table.
来源：Model versions and lifecycle ↗官方变更公告记录 JSON Output、Function Calling、Chat Prefix Completion 和 FIM Completion 等能力。
来源：DeepSeek API Upgrade ↗Anthropic 官方公告明确 Claude 3.5 Sonnet 可通过 Anthropic API 使用，同时提供 Claude.ai、iOS、Amazon Bedrock 和 Vertex AI 渠道。
来源：Claude 3.5 Sonnet ↗Google Cloud lists gemini-1.5-pro-001 release date as May 24, 2024 in the Gemini Enterprise Agent Platform lifecycle table.
来源：Model versions and lifecycle ↗Anthropic 官方公告发布 Claude 3 Haiku、Claude 3 Sonnet 和 Claude 3 Opus 三个模型。
来源：Introducing the next generation of Claude ↗Google 官方公告介绍 Gemini，并区分 Ultra、Pro 和 Nano 三种规模。
来源：Introducing Gemini: our largest and most capable AI model ↗Google DeepMind 官方历史页说明 DeepMind started in 2010。本事件记录组织前身起源，不等同于 Google DeepMind 合并日。
来源：About Google DeepMind ↗百川官方 README 记录 Baichuan-7B 的模型介绍和公开版本，但该页面未提供可核验的发布日期。
来源：baichuan-inc/Baichuan-7B ↗DeepSeek 官网原文称“DeepSeek-V4-Flash 正式版 API 已上线公测”，并强调 Agent 能力大幅增强；V4 Pro 暂未变动。
来源：DeepSeek official website ↗Google's deprecations table lists July 21, 2026 as the release date for gemini-3.6-flash.
来源：Gemini deprecations ↗Anthropic 官方 API 生命周期页记录 Claude Opus 4.1 于 2026-06-05 标记 Deprecated，计划不早于 2026-08-05 退休，推荐迁移到 Claude Opus 4.8。
来源：Model deprecations | Claude API ↗DeepSeek 官方公告记录 V4 Preview 于 2026-04-24 上线并开源，明确包含 V4 Pro 与 V4 Flash，API 当日可用。
来源：DeepSeek-V4 Preview Release ↗Anthropic 官方 API 生命周期页记录 Claude Haiku 3 于 2026-04-20 退休，推荐迁移到 Claude Haiku 4.5。
来源：Model deprecations | Claude API ↗Google lists December 2, 2025 as the earliest possible shutdown date and recommends gemini-3.1-pro-preview.
来源：Gemini deprecations ↗Google lists December 2, 2025 as the earliest possible shutdown date and recommends gemini-3.1-pro-preview.
来源：Gemini deprecations ↗Google Cloud lists September 24, 2025 as the retirement date and gemini-2.5-flash as the recommended upgrade. The statement is scoped to the cited Gemini Enterprise Agent Platform lifecycle table.
来源：Model versions and lifecycle ↗NVIDIA 发布基于 Llama 3.3 70B Instruct、面向推理和 Agent 场景的 Nemotron Super 版本。
来源：Llama-3.3-Nemotron-Super-49B-v1.5 model card ↗百度官方 ERNIE 仓库记录 ERNIE 4.5 于 2025 年 6 月发布；本事件按月精度保存，不推断具体日。
来源：PaddlePaddle/ERNIE ↗Google Cloud lists May 24, 2025 as the retirement date and gemini-2.5-flash as the recommended upgrade. The statement is scoped to the cited Gemini Enterprise Agent Platform lifecycle table.
来源：Model versions and lifecycle ↗TII 发布采用 Transformer 与 Mamba-2 SSM 混合架构的 Falcon-H1 7B 指令模型。
来源：Falcon-H1-7B-Instruct model card ↗该事件只记录来源发布行为。 Harvey's public BigLaw Bench evaluation names Gemini 2.5 Pro Preview and reports an 85.02% score. The described legal workflows include transactional due diligence, transaction structuring, litigation drafting, document review and analysis, with reasoning over inputs consisting of hundreds of pages.
来源：Harvey: validating Gemini 2.5 Pro Preview's advanced legal reasoning with BigLaw Bench ↗Skywork 官方仓库记录 OR1-32B 与 OR1-7B 最终版本于 2025-05-13 发布。
来源：SkyworkAI/Skywork-OR1 ↗ByteDance Seed 官方仓库记录 Seed1.5-VL 以 doubao-1-5-thinking-vision-pro-250428 模型 ID 于 2025-05-12 在火山引擎发布。
来源：ByteDance-Seed/Seed1.5-VL ↗NVIDIA 发布基于 Llama 3.1 405B Instruct、经过 NAS 和知识蒸馏优化的 Nemotron Ultra 推理模型。
来源：Llama-3.1-Nemotron-Ultra-253B-v1 model card ↗NVIDIA 发布面向效率和本地部署的 Nemotron Nano 推理模型。
来源：Llama-3.1-Nemotron-Nano-8B-v1 model card ↗Microsoft published Phi-4 as a 14B MIT-licensed open-weight model.
来源：Phi-4 model card ↗官方公告介绍 Context Caching on Disk，并说明可降低重复输入的延迟和 API 成本；具体价格以当前价格页为准。
来源：DeepSeek API introduces Context Caching on Disk ↗DeepSeek 官网当前明确展示 Responses API 公测入口；官方页面未公开该入口的发布日期，因此日期保持未知。
来源：DeepSeek official website ↗Meta 官方 Llama Models 页面当前展示 Llama 4 多模态模型及其产品信息；页面未公开该目录入口的发布日期，因此日期保持未知。
来源：Llama models ↗该事件只记录来源发布行为。 Google Cloud's public case names Gemini 1.5 Pro in 7-Eleven Vietnam's internal IT chatbot, built with Vertex AI Agent Builder and Dialogflow CX. The chatbot supports employees across 140 stores, resolves printer and other IT questions with structured flows and documentation, and the case reports 50% less time spent fixing IT issues. The team completed proof-of-concept and testing before production.
来源：7-Eleven Vietnam: Easing IT workloads with a customized, AI-powered chatbot ↗该事件只记录来源发布行为。 ASAPP says its GenerativeAgent achieved a 25-40% improvement in core business outcomes compared with other AI models in production after switching to Claude in Amazon Bedrock; the exact Claude version is not published.
来源：ASAPP Claude Platform (API) case study ↗该事件只记录来源发布行为。 Mistral says initial ASML use cases include AI agents for software reliability, early issue detection, log analysis for root-cause identification and technical product-documentation automation.
来源：ASML and Mistral ↗该事件只记录来源发布行为。 Assembled explicitly identifies Claude 3 Haiku as the high-speed operations model in Assembled Assist. The case reports platform and customer support outcomes including higher CSAT, fewer escalations, more cases solved per hour, higher case automation, and customer examples from Thrasio and Honeylove.
来源：Assembled Claude Platform (API) case study ↗该事件只记录来源发布行为。 Mistral says BMW is developing industry-specific AI models for crash simulation using BMW proprietary datasets and engineering expertise, targeting quality, accuracy and speed improvements for complex engineering tasks.
来源：BMW and Mistral ↗该事件只记录来源发布行为。 Mistral says AI use cases assist cargo release, fact-checking, client information requests and shipping-document processing, targeting faster and higher-quality responses and improved customer-facing processes.
来源：CMA CGM and Mistral ↗该事件只记录来源发布行为。 Google Cloud's public case names Gemini 1.5 Pro in Gelato's engineering-support ticket triage and customer-support error categorization. It reports 120 hours of weekly manual tasks saved and ticket-assignment accuracy increasing from 60% to 90%; Gemini automatically assigns tickets to the responsible engineering team. Gelato also describes instant error classification in customer-support systems.
来源：Gelato: Bringing efficient, high-quality local production to global ecommerce with Gemini ↗该事件只记录来源发布行为。 Mistral says the partnership equips HSBC colleagues with tools to innovate, simplify daily tasks and free up time to deliver for customers.
来源：HSBC and Mistral ↗该事件只记录来源发布行为。 Hershey Entertainment & Resorts uses Google Cloud Customer Engagement Suite, Conversational Agents, Vertex AI, Gemini and connected enterprise systems to automate guest and employee support. The public case reports chatbot launch in less than 40 hours, more than 25% of employee call-off requests automated, 50%+ of password-reset calls managed, and eight hours of employee work saved per day.
来源：Hershey Entertainment & Resorts: Gifting all guests with a sweet time using Customer Engagement Suite ↗该事件只记录来源发布行为。 Square Yards uses Gemini Enterprise Agent Platform and Gemini APIs with other Google Cloud services to run SuperAgent Pro, a multilingual conversational voice engine. The public case reports 200,000 automated calls per day, sub-700ms conversational response latency, 12+ Indian languages and 10 billion Gemini AI tokens consumed per month.
来源：Square Yards uses Google Cloud and Gemini to scale in-house agentic AI, running 200,000 daily calls with sub-700ms latency ↗