MAAS / WEEKLY
DAILY DIFF · SIGNAL FEED

每日动态

自动抓取 16 个平台 + 行业信源，与上一次快照逐行对比。已过滤页面抖动噪声，按"实质变化 / 常规波动"分层展示；每日要点由 AI 从增量中提炼。数据每天更新。

最新

2026-09-04

实质变化 20成功 61失败 6共 67 个信源
AI 今日要点
  • 发布OpenAIOpenAI 发布旗舰模型 GPT-6 Astra，面向企业 Trusted Access Program 开放 API 访问，定价 $10/MTok 输入、$50/MTok 输出，并新增 Async tool calling、Mid-turn steering 等控制功能。
  • 发布Google Vertex AIGoogle Vertex AI 上线 Gemini 3.8 Flash，主打长时编码与自主智能体场景，并推出 $0.75/$3.75 每百万 tokens 的入门定价及 50% Provisioned Throughput 促销信用。
  • 动态硅基流动硅基流动新增 zai-org/GLM-5.3、MiniMax-M2.5、Qwen3.5-397B-A17B、Nex-N2-Pro 等模型，同时宣布将于 2026-09-11 下线 Nex-N2-Pro、Qwen3.5-397B-A17B、MiniMax-M2.5 等模型。
  • 调价硅基流动硅基流动调整 DeepSeek-V4-Flash 分时段定价及 DeepSeek-V4-Pro 缓存命中输入 tokens 价格，并宣布 /user/info 接口将停止服务。
  • 调价Google Vertex AIGoogle Vertex AI 新增 Gemini 3.5 Live Translate 定价（音频输入 $3.50/百万、输出 $21.00/百万），并扩展 Gemini 3.8 Flash 的音频 token 计费说明。
  • 动态LMArenaLMArena 排行榜新增 claude-fable-5.1-max（Elo 1765）、hy4-preview（1626）等模型，其中 claude-fable-5.1-max 大幅领先其他新上榜模型。
火山方舟定价+0 / -4来源 ↗
删除 4 行
  • SDK 升级与使用说明
  • 知识型视频智能问答解决方案
  • 视频高光提取及检索解决方案
  • 音视频知识问答核心流程
火山方舟更新日志+2 / -0来源 ↗
Computer Use Agent 服务规则
新增 2 行
  • Computer Use Agent 服务规则
  • 用量明细 CSV 解析最佳实践
硅基流动定价+1 / -2来源 ↗
zai-org/GLM-5.3
新增 1 行
  • zai-org/GLM-5.3
删除 2 行
  • GLM-Z1-9B-0414
  • THUDM/GLM-Z1-9B-0414
硅基流动更新日志~1 处变更+6 / -0来源 ↗
MiniMaxAI/MiniMax-M2.5
变更 1 处（旧 → 新）
  • 【模型价格调整】DeepSeek-V4-Flash 模型分时段定价调整【接口服务调整】/user/info 接口将停止服务【模型价格调整】DeepSeek-V4-Pro 缓存命中输入 tokens 价格调整【模型价格调整】Nex-N2-Pro、DeepSeek-V4-Pro、DeepSeek-V3.2、Qwen3.6 →【模型服务调整】Nex-N2-Pro、Qwen3.5-397B-A17B、MiniMax-M2.5 等模型将下线【模型价格调整】DeepSeek-V4-Flash 模型分时段定价调整【接口服务调整】/user/info 接口将停止服务【模型价格调整】DeepSeek-V4-Pro 缓存命中输入 tokens 价格调整【
新增 6 行
  • MiniMaxAI/MiniMax-M2.5
  • Pro/MiniMaxAI/MiniMax-M2.5
  • Qwen/Qwen3.5-397B-A17B
  • nex-agi/Nex-N2-Pro
  • 【模型服务调整】Nex-N2-Pro、Qwen3.5-397B-A17B、MiniMax-M2.5 等模型将下线
  • 为进一步优化资源配置，提供更先进、优质的技术服务，平台将于 2026-09-11 对下列模型进行下线处理：
OpenAI模型列表~2 处变更+13 / -2来源 ↗
$10 / Input MTok
变更 2 处（旧 → 新）
  • If you're not sure where to start, use GPT-5.6 Sol, our flagship model for complex reasoning and coding. Choose GPT-5.6 Terra to balance intelligence and cost, →If you're not sure where to start, use GPT-6 Astra, our flagship model for complex reasoning and coding. Choose GPT-5.6 Terra to balance intelligence and cost,
  • Using GPT-5.6→Using GPT-6 Astra
新增 13 行
  • $10 / Input MTok
  • $50 / Output MTok
  • Async tool calling
  • Compare model capabilities and specifications.
  • GPT-6 Astra
  • GPT‑6 Astra is rolling out today for enterprises in our Trusted Access Program⁠, with access through API and our Plus, Pro, Business and Enterprise plans coming in the coming days.
  • Mid-turn steering
  • Misalignment monitoring
  • Our most capable model, built for the hardest end-to-end work
  • Safety classifiers
  • Under-18 guidance
  • gpt-6-astra
  • lowmediumhighxhighmax
删除 2 行
  • Start with GPT-5.6 Sol for complex reasoning and coding, choose GPT-5.6 Terra to balance intelligence and cost, or use GPT-5.6 Luna for cost-sensitive, high-volume workloads.
  • Under 18 API Guidance
OpenAI定价~1 处变更+10 / -1来源 ↗
Fast mode is unavailable for GPT-6 Astra with EU data residency. Use Standard processing for those requests. See Fast mode compatibility.
变更 1 处（旧 → 新）
  • Using GPT-5.6→Using GPT-6 Astra
新增 10 行
  • Async tool calling
  • Fast mode is unavailable for GPT-6 Astra with EU data residency. Use Standard processing for those requests. See Fast mode compatibility.
  • GPT‑6 Astra is rolling out today for enterprises in our Trusted Access Program⁠, with access through API and our Plus, Pro, Business and Enterprise plans coming in the coming days.
  • Mid-turn steering
  • Misalignment monitoring
  • Safety classifiers
  • Under-18 guidance
  • | gpt-6-astra | $10.00 | $1.00 | $12.50 | $50.00 | $20.00 | $2.00 | $25.00 | $75.00
  • | gpt-6-astra | $20.00 | $2.00 | $25.00 | $100.00 | $40.00 | $4.00 | $50.00 | $150.00
  • | gpt-6-astra | $5.00 | $0.50 | $6.25 | $25.00 | $10.00 | $1.00 | $12.50 | $37.50
删除 1 行
  • Under 18 API Guidance
OpenAI更新日志~1 处变更+22 / -1来源 ↗
Added new controls for long-running work with GPT-6 Astra in the Responses API:
变更 1 处（旧 → 新）
  • Using GPT-5.6→Using GPT-6 Astra
新增 22 行
  • Added new controls for long-running work with GPT-6 Astra in the Responses API:
  • Async tool calling
  • Async tool calling: Let the model continue working while your application runs function or custom tools, then return results as they become available.
  • Change reasoning effort mid-conversation: Increase effort for difficult work or reduce it for routine follow-ups while preserving the cached prompt prefix.
  • Connections to api.openai.com can now use IPv6.
  • GPT-6 Astra does not support custom temperature or top_p values or log probabilities (logprobs).
  • GPT-6 Astra does not support the none reasoning effort level.
  • Key changes to consider when migrating:
  • Mid-turn steering
  • Mid-turn steering: Send additional instructions while a response is in progress over WebSockets, so the model can incorporate corrections or changing requirements.
  • Misalignment monitoring
  • Misalignment monitoring asynchronously checks for potential issues during agent work in supported Responses API requests. Checks can trigger safety alerts or stop a conversation for review.
  • Released GPT-6 Astra, our most capable model, built for the hardest end-to-end work.
  • Safety classifiers
  • September, 2026
删除 1 行
  • Under 18 API Guidance

展示前 15 条新增 / 8 条删除，完整数据见仓库 data/diff/ 目录

Anthropic模型列表+1 / -1来源 ↗
新增 1 行
  • | Show all details
删除 1 行
  • | Show all detailsShow fewer details
Anthropic更新日志~1 处变更+1 / -0来源 ↗
The guides for the Claude Enterprise endpoints of the Admin API (user management and spend limits), the Claude Enterprise Analytics API, and the Compliance API now show the anthropic-version header; send it on every request to these endpoints, as in the rest of the Claude API. See API versions.
变更 1 处（旧 → 新）
  • Text generated by Claude Fable 5.1 and Claude Mythos 5.1 carries Anthropic's text watermark, and supported image and video files that Claude produces through th→Text generated by Claude Fable 5.1 and Claude Mythos 5.1 carries Anthropic's text watermark, and supported image, video, and audio files that Claude produces th
新增 1 行
  • The guides for the Claude Enterprise endpoints of the Admin API (user management and spend limits), the Claude Enterprise Analytics API, and the Compliance API now show the anthropic-version header; send it on every requ
Google Gemini API博客+1 / -1来源 ↗
5 amazing visuals show how the male fruit fly’s brain map is advancing neuroscience
新增 1 行
  • 5 amazing visuals show how the male fruit fly’s brain map is advancing neuroscience
删除 1 行
  • AMIE, our research medical AI system, demonstrates real-time clinical video consultation capabilities in a first-of-its-kind study.
Google Vertex AI模型列表~2 处变更+5 / -2来源 ↗
Built for long-horizon coding and autonomous agents
变更 2 处（旧 → 新）
  • Last updated 2026-09-01 UTC.→Last updated 2026-09-02 UTC.
  • [[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Hard to understand"→[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Hard to understand"
新增 5 行
  • Built for long-horizon coding and autonomous agents
  • Gemini 3.8 Flash
  • Improved response quality
  • Our intelligent everyday driver for developers, delivering a step forward in software engineering and feeling distinctly better to build with.
  • Our most intelligent workhorse model yet, built for long-horizon coding and autonomous agents.
删除 2 行
  • Enhanced efficiency and practical reasoning, helping you build and iterate with greater ease
  • Greater token efficiency
Google Vertex AI定价~7 处变更+22 / -3来源 ↗
$2.00 / 1M tokens |
变更 7 处（旧 → 新）
  • Effective August 13, 2026 through December 31, 2026, Google Cloud is injecting a monthly billing credit equal to 50% of net eligible Provisioned Throughput spen→Effective August 13, 2026 through December 31, 2026, Google Cloud is injecting a monthly billing credit equal to 50% of net eligible Provisioned Throughput spen
  • Gemini 3.7 Flash, Gemini 3.6 Flash, and CodeMender using these models are offered with introductory pricing of $0.75 / $3.75 per 1M tokens input / output throug→Gemini 3.5 Live Translate
  • Global | $3.00 | $3.00 |→Global | $0.375 | $0.375 | $0.0375 | $0.0375
  • Global | - |→Global | $1.35 | $1.35 | $0.135 | $0.135
  • Image output*** | - |→Image output*** | $30.00 |
  • Input (text, image)*** | - |→Input (text, image)*** | $0.30 |
  • Issued credits apply to usage across On-Demand Gemini 3.6 Flash and 3.7 Flash SKUs, as well as Provisioned Throughput SKUs across all models. Credits are issued→Issued credits apply to usage across On-Demand Gemini 3.8 Flash, Gemini 3.7 Flash and Gemini 3.6 Flash SKUs, as well as Provisioned Throughput SKUs across all m
新增 22 行
  • $2.00 / 1M tokens |
  • * Billing is based on total input and output audio token consumption, calculated at a rate of 25 tokens per second of audio.
  • 50% Promotional Credits on Gemini 3.8 Flash, Gemini 3.7 Flash and Gemini 3.6 Flash Provisioned Throughput Usage
  • Audio Input | $3.50 / 1,000,000 count
  • Audio Output | $21.00 / 1,000,000 count
  • Claude Fable 5.1 |
  • Gemini 3.5 Live Translate |
  • Gemini 3.8 Flash
  • Gemini 3.8 Flash*
  • Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash, and CodeMender using these models are offered with introductory pricing of $0.75 / $3.75 per 1M tokens input / output through December 31, 2026. Starting January 1, 2
  • Global | $1.875 | $1.875 |
  • Global | $6.75 | $6.75 |
  • Low-latency, real-time speech to speech translation model that supports 70+ languages. |
  • Non-global | $0.4125 | $0.4125 | $0.04125 | $0.04125
  • Non-global | $0.825 | $0.825 | $0.0825 | $0.0825
删除 3 行
  • $0.000001 / 1 hour |
  • $2.50 / 1M tokens |
  • 50% Promotional Credits on Gemini 3.6 Flash & Gemini 3.7 Flash Provisioned Throughput Usage

展示前 15 条新增 / 8 条删除，完整数据见仓库 data/diff/ 目录

Google Vertex AI博客+2 / -3来源 ↗
By Nick Galloway • 3-minute read
新增 2 行
  • By Nick Galloway • 3-minute read
  • Getting started with Mantis, our open-source bug finding-and-fixing harness
删除 3 行
  • By Sergio Villani • 19-minute read
  • Developers & Practitioners
  • 10 questions every startup should answer before moving to production with their AI prototype
Cohere定价+3 / -3来源 ↗
$4,300Pricing | Secure and Scalable Enterprise AI | Cohere
新增 3 行
  • $4,300Pricing | Secure and Scalable Enterprise AI | Cohere
  • Retrieval optimization model
  • Semantic representation model
删除 3 行
  • Pricing | Secure and Scalable Enterprise AI | Cohere
  • Search and discovery model
  • Semantic search ranking
Cohere博客+6 / -6来源 ↗
Automation’s Early Footprint
新增 6 行
  • Automation’s Early Footprint
  • Enterprise AIFor Business
  • How small AI models can make a big impact for enterprises
  • Retrieval optimization model
  • Semantic representation model
  • Where AI Agents Are (and Aren’t) Being Built
删除 6 行
  • A high-throughput vision parsing model with the strongest price-performance profile on the market.
  • Cohere triples UK footprint with new London office to support R&D growth
  • LLM Serving Fairness: No more noisy neighbors
  • Product LaunchCompany NewsEnterprise AI
  • Search and discovery model
  • Semantic search ranking
xAI模型列表+1 / -0来源 ↗
Imagine Image Quality Retirement on Nov 2
新增 1 行
  • Imagine Image Quality Retirement on Nov 2
xAI定价+1 / -0来源 ↗
Imagine Image Quality Retirement on Nov 2
新增 1 行
  • Imagine Image Quality Retirement on Nov 2
HuggingFace行业数据+3 / -3来源 ↗
"base_model:finetune:runwayml/stable-diffusion-v1-5",
新增 3 行
  • "base_model:finetune:runwayml/stable-diffusion-v1-5",
  • "base_model:runwayml/stable-diffusion-v1-5",
  • "license:creativeml-openrail-m",
删除 3 行
  • "arxiv:2309.00071",
  • "base_model:Qwen/Qwen3-8B-Base",
  • "base_model:finetune:Qwen/Qwen3-8B-Base",
Artificial Analysis行业数据+19 / -0来源 ↗
All reasoning and effort variants of each selected release · Weighted average cost (USD) per Artificial Analysis Intelligence Index task
新增 19 行
  • All reasoning and effort variants of each selected release · Weighted average cost (USD) per Artificial Analysis Intelligence Index task
  • AnthropicMetaOpenAISpaceXAIKimiZ AIGoogleAlibabaDeepSeekMBZUAI Institute of Foundation ModelsMiniMaxThinking MachinesMistralCohere
  • AnthropicMetaOpenAISpaceXAIKimiZ AIGoogleDeepSeekNVIDIAOther Models
  • Arena Elo: average Elo rating of the model · Higher is better
  • Artificial Analysis Finance & Accounting Index
  • Benchmarking GPT-6 Astra
  • Claude Fable 5.1 (Adaptive Reasoning, Low Effort, Default Fallback)See more
  • Finance & AccountingStrategy & OpsLegalHealthcare & MedicalEngineeringEconomics
  • GPT-6 Astra (Non-reasoning)
  • GPT-6 Astra (high)
  • GPT-6 Astra (low)
  • GPT-6 Astra (max)
  • GPT-6 Astra (medium)
  • GPT-6 Astra (xhigh)
  • Gemini 3.8 Flash (high)

展示前 15 条新增 / 8 条删除，完整数据见仓库 data/diff/ 目录

LMArena行业数据~2 处变更+17 / -7来源 ↗
Bytedancedreamina-seedance-2.5-720p: 1478±10
变更 2 处（旧 → 新）
  • 4Bytedancedreamina-seedance-2.0-720p1477±8→5Bytedancedreamina-seedance-2.0-720p1477±8
  • claude-opus-5-high1661+8/-8→claude-opus-5-high1661+7/-7
新增 17 行
  • 4Bytedancedreamina-seedance-2.5-720p1478±10
  • claude-fable-5.1-max |
  • claude-fable-5.1-max1504±11
  • claude-fable-5.1-max1765+23/-23
  • claude-fable-51507±5
  • claude-opus-5-high1493±5
  • flux-3-video-202608111449±6
  • gemini-3.8-flash-high |
  • gemini-3.8-flash-high1494
  • gemini-omni-flash1462±6
  • grok-imagine-video-1.5-720p1456±5
  • hy4-preview1626
  • minimax-h31497±6
  • muse-spark-1.11492±5
  • qwen3.8-max-09021688
删除 7 行
  • 3Bytedancedreamina-seedance-2.5-720p1483±12
  • claude-fable-51508±5
  • claude-opus-5-high1492±5
  • flux-3-video-202608111447±6
  • gemini-3.7-flash-high1491
  • gemini-omni-flash1463±6
  • gpt-5.6-sol-xhigh (codex-harness)1616+7/-7

展示前 15 条新增 / 8 条删除，完整数据见仓库 data/diff/ 目录

◌ 1 个信源仅有常规波动Anthropic
  • Anthropic · 博客 — 数据抖动 2 行，已过滤
⚠ 6 个信源抓取失败
  • 火山方舟 · 模型列表 — 提取后内容过短 (raw_len=11071)
  • 腾讯混元/TokenHub · 模型列表 — 提取后内容过短 (raw_len=29182)
  • 腾讯混元/TokenHub · 定价 — 提取后内容过短 (raw_len=29182)
  • 腾讯混元/TokenHub · 更新日志 — 提取后内容过短 (raw_len=29182)
  • 腾讯混元/TokenHub · 博客 — 提取后内容过短 (raw_len=29182)
  • OpenAI · 博客 — 提取后内容过短 (raw_len=9951)
2

2026-09-02

实质变化 0成功 57失败 10共 67 个信源
当日无变化（57 个信源首次抓取建立基线）
⚠ 10 个信源抓取失败
  • 腾讯混元/TokenHub · 模型列表 — 提取后内容过短 (raw_len=29182)
  • 腾讯混元/TokenHub · 定价 — 提取后内容过短 (raw_len=29182)
  • 腾讯混元/TokenHub · 更新日志 — 提取后内容过短 (raw_len=29182)
  • 腾讯混元/TokenHub · 博客 — 提取后内容过短 (raw_len=29182)
  • OpenAI · 博客 — 提取后内容过短 (raw_len=9951)
  • Google Gemini API · 模型列表 — curl 返回空 (stderr: )
  • Google Gemini API · 定价 — curl 返回空 (stderr: )
  • Google Gemini API · 更新日志 — curl 返回空 (stderr: )
  • xAI · 更新日志 — 提取后内容过短 (raw_len=1383)
  • xAI · 博客 — 提取后内容过短 (raw_len=1383)

数据由 fetch_sources.py 每日自动抓取生成，经 diff_clean 降噪过滤与 LLM 要点提炼 · 最近同步: 2026-09-04 18:06:59 · 完整历史快照见仓库 data/ 目录