模型、公司、政策与基础设施层面的行业动向。（本日 3 条）
OpenAI说明安全治理和平台控制变化 一般 topic|multi_source
OpenAI News RSS / OpenAI Blog RSS / OpenAI Company News RSS
follow-builders X发布 Claude Code agent 工具工作流 一般 topic|primary
follow-builders X feed
AWS更新agent 与开发者工具能力 一般 topic|primary
AWS Machine Learning BlogAIGC 内容生产与创作工作流相关动态。（本日 1 条）
NVIDIA更新agent 与开发者工具能力 一般 topic|primary
NVIDIA Newsroom RSS面向开发者的产品、工具与工程实践更新。（本日 3 条）
OpenAI News RSS: Omio 一般 topic|multi_source
OpenAI News RSS / OpenAI Blog RSS / OpenAI Company News RSS
Alibaba Cloud更新agent 与开发者工具能力 一般 topic|primary
Alibaba Cloud Blog
Official Microsoft Blog: Powering The Next Wave Of AI Expanding Capacity With Our New Datacenter In Pecos 一般 topic|primary
Official Microsoft Blog值得关注的开源模型与项目动态。（本日 1 条）
GitHub Changelog更新agent 与开发者工具能力 一般 topic|primary
GitHub Changelog
NVIDIA披露 agent 与开发者工具能力NVIDIA Developer Blog / 2026-06-23NVIDIA更新agent 工作流和开发工具能力，重点落在任务编排、上下文、权限控制、工程集成和失败恢复。更有价值的信息是agent 工作流、开发工具入口、权限控制和工程集成，判断这类方案时还要看落地质量取决于权限模型、评估回放、团队流程和可观测性。文章拆解 agent、开发工具或自动化流程里的任务规划、权限、上下文、工具调用和失败恢复。
DeusData/codebase-memory-mcp 值得关注 NEW 本周 +7,560 stars: 领域：agent 适合：适合先评估该仓库的运行入口、集成边界、许可证和近期维护，再决定是否进入团队试用。；#1，较前一日持平 calesthio/OpenMontage 值得关注 NEW 本周 +6,089 stars: 领域：agent 适合：适合先评估该仓库的运行入口、集成边界、许可证和近期维护，再决定是否进入团队试用。；#2，较前一日持平 Panniantong/Agent-Reach 值得关注 ↑ UP +5 本周 +8,108 stars: 领域：agent 适合：适合先评估该仓库的运行入口、集成边界、许可证和近期维护，再决定是否进入团队试用。；#3，较前一日上升 5 位 n0-computer/iroh 一般 NEW 本周 +1,806 stars: 领域：AI tooling 适合：适合先评估该仓库的运行入口、集成边界、许可证和近期维护，再决定是否进入团队试用。；#4，较前一日持平 google-research/timesfm 一般 NEW 本周 +4,259 stars: 领域：AI tooling 适合：适合先评估该仓库的运行入口、集成边界、许可证和近期维护，再决定是否进入团队试用。；#5，较前一日持平 iptv-org/iptv 一般 ↓ DOWN -1 本周 +5,017 stars: 领域：AI tooling 适合：适合先评估该仓库的运行入口、集成边界、许可证和近期维护，再决定是否进入团队试用。；#6，较前一日下降 1 位 asgeirtj/system_prompts_leaks 一般 ↑ UP +2 本周 +2,612 stars: 领域：agent 适合：适合先评估该仓库的运行入口、集成边界、许可证和近期维护，再决定是否进入团队试用。；#7，较前一日上升 2 位 withastro/flue 一般 NEW 本周 +1,413 stars: 领域：agent 适合：适合先评估该仓库的运行入口、集成边界、许可证和近期维护，再决定是否进入团队试用。；#8，较前一日持平
OpenRouter| Rank | Model | Provider | Value | Change |
|---|---|---|---|---|
| #1 | DeepSeek V4 Flash | deepseek | 4.96T tokens | 16% |
| #2 | MiMo-V2.5 | xiaomi | 4.1T tokens | 12% |
| #3 | MiniMax M3 | minimax | 3.77T tokens | 17% |
| #4 | Hy3 preview | tencent | 3.68T tokens | 5% |
| #5 | Owl Alpha | openrouter | 2.7T tokens | 10% |
| #6 | DeepSeek V4 Pro | deepseek | 2.5T tokens | 32% |
| #7 | Claude Opus 4.7 | anthropic | 2.45T tokens | 3% |
| #8 | Claude Opus 4.8 | anthropic | 1.73T tokens | 39% |
| #9 | Claude Sonnet 4.6 | anthropic | 1.53T tokens | 24% |
| #10 | GLM 5.2 | z-ai | 1.51T tokens | new |
Artificial Analysis| Rank | Model | Provider | Value | Change |
|---|---|---|---|---|
| #1 | Claude Fable 5 (with fallback) | anthropic | 60 分 | AA Index |
| #2 | Claude Opus 4.8 (max) | anthropic | 56 分 | AA Index |
| #3 | GPT-5.5 (xhigh) | openai | 55 分 | AA Index |
| #4 | Claude Opus 4.7 (max) | anthropic | 54 分 | AA Index |
| #5 | GPT-5.5 (high) | openai | 53 分 | AA Index |
| #6 | GLM-5.2 (max) | zhipu | 51 分 | AA Index |
| #7 | Gemini 3.5 Flash | 50 分 | AA Index | |
| #8 | Claude Sonnet 4.6 (max) | anthropic | 47 分 | AA Index |
| #9 | Gemini 3.1 Pro Preview | 46 分 | AA Index | |
| #10 | Qwen3.7 Max | alibaba | 46 分 | AA Index |
SWE-bench Pro| Rank | Model | Provider | Value | Change |
|---|---|---|---|---|
| #1 | gpt-5.4 (xHigh)* | openai | 59.10±3.56% | Resolve Rate |
| #2 | Muse Spark* | scale | 55.00±3.60% | new |
| #3 | claude-opus-4-6 (thinking)* | anthropic | 51.90±3.61% | Resolve Rate |
| #4 | gemini-3.1-pro (thinking)* | 46.10±3.60% | Resolve Rate | |
| #5 | claude-opus-4-5-20251101 | anthropic | 45.89±3.60% | Resolve Rate |
| #6 | claude-4-5-Sonnet | anthropic | 43.60±3.60% | Resolve Rate |
| #7 | gemini-3-pro-preview | 43.30±3.60% | Resolve Rate | |
| #8 | claude-4-Sonnet | anthropic | 42.70±3.59% | Resolve Rate |
| #9 | gpt-5-2025-08-07 (High) | openai | 41.78±3.49% | Resolve Rate |
| #10 | gpt-5.2-codex | openai | 41.04±3.57% | Resolve Rate |
NVIDIA披露模型评估和研究结果NVIDIA Newsroom RSS / 2026-06-22NVIDIA披露模型能力和评估方法更新，重点落在能力边界、评估设置、数据来源、使用场景和限制说明。更有价值的信息是模型能力、评估设置、数据来源和限制说明，判断这类方案时还要看结论仍要依赖可复现评测、真实任务和公开限制。文章梳理模型评测或研究结论怎样改变能力边界、成本预期和可靠性判断。
57场面试杀进OpenAI！华人博士开源「AI面经」，含泪推荐QbitAI / 2026-06-2357场面试杀进OpenAI！华人博士开源「AI面经」，含泪推荐：希望你能找到快乐。这条 RSS 适合观察国内 AI 产品、产业反馈和工程实践的讨论方向；读者若要引用事实、数据或公司动作，仍应点开原文并继续追溯一手来源核对。
Thibault Sottiaux@thsottiauxLet's Patch The Planet. Updates to codex security and a new GPT-5.5-Cyber. A day of celebration for cyber defense acceleration. https://t.co/SYKQvsTA44
Aaron Levie@levieAlmost all AI model and agent progress is downstream from evals. Open weights post training for specific domains comes down to evals. Agent improvements in the applied AI layer is all about evals. Agentic enterprise d...
Swyx@swyxi dont think anyone is correctly doing the math around how SpaceX, the NeoCloud+NeoLab, is currently going to market? SpaceX has already recouped about HALF its investment in Cursor, in compute deals. The other half i...
Peter Yang@petergyangI want to do a podcast episode with someone who's good at using Codex / Claude Code to create fun pixel or threejs games and can show us how to do it. I am a gamer at heart! Who's the best person to talk to?
Guillermo Rauch@rauchgClaude Design → Vercel, in one click https://t.co/Btq9hFk7OB https://t.co/NpgdokzpvE
Aaron Levie@levieWe heard that HTML is a big deal again. You can now preview, edit, manage versions, and securely share any HTML based content on Box. Great for being able to work with any agent produced content immediately. https://t...
Ryo Lu@ryolu_here's my talk at Cursor Compile some thoughts on how we build in the age of AI and what doesn't change https://t.co/otMw3nwPkr
