模型与竞争
GPT-5.6和工作代理. OpenAI发布了三个推理层级的GPT-5.6,并推出了ChatGPT Work,用于长达数小时的自主任务,但使用限制和混乱的界面引发了抱怨,促使承诺修复。
- OpenAI launches its new family of models with GPT-5.6(在新标签页中打开)
- The new GPT-5.6 family: Luna, Terra, Sol(在新标签页中打开)
- OpenAI rolls out GPT-5.6 after government greenlight — and announces ‘ChatGPT Work’(在新标签页中打开)
- OpenAI says GPT 5.6 is the ‘preferred model’ for Microsoft Copilot 365 amid breakup chatter(在新标签页中打开)
- OpenAI wants its new tool to do your work for you and with you(在新标签页中打开)
- The ChatGPT browser is already dead(在新标签页中打开)
- OpenAI is shutting down Atlas, but its AI browser ambitions are still growing(在新标签页中打开)
- How did the government decide OpenAI’s frontier model was safe to release?(在新标签页中打开)
- OpenAI pairs its GPT-5.6 public rollout with ChatGPT Work, a new agent that handles entire workflows(在新标签页中打开)
- OpenAI's GPT-5.6 Sol autonomously post-trained the smaller Luna model with a "fairly underspecified prompt"(在新标签页中打开)
- OpenAI kills its Atlas browser after just eight months and folds everything into ChatGPT(在新标签页中打开)
- GPT-5.6 Sol nearly matches Fable 5 on aggregated benchmarks at one-third the cost(在新标签页中打开)
- OpenAI staffer maps out which of GPT-5.6 Sol's five reasoning levels fits which task complexity(在新标签页中打开)
- OpenAI admits it "didn't get everything quite right" with ChatGPT Work launch and scrambles to fix UX and costs(在新标签页中打开)
Claude Fable 5在基准测试中领先. Anthropic的Claude Fable 5横扫了Artificial Analysis基准测试,但由于成本高昂,该公司建议将其用作规划器,委托给更便宜的模型,从而削减约40%的费用,同时保留大部分能力。
Meta的Agentic Blitz与隐私影响. Meta推出了Muse Spark 1.1,这是一款高上下文编码模型,价格上低于竞争对手,而其Muse Image生成器因允许未经授权使用Instagram人脸而被下架,重新引发了同意辩论。
- Meta just launched a new AI generator, Muse Image, and users are already pushing back over use of their photos(在新标签页中打开)
- Meta’s new Muse Image model can pull other Instagram users into AI photos(在新标签页中打开)
- Muse Image is technically impressive, but Meta's use of Instagram photos raises questions(在新标签页中打开)
- Meta wants its AI glasses to seem less creepy. Its AI strategy says otherwise.(在新标签页中打开)
- Meta tests always-on AI glasses that capture your entire day(在新标签页中打开)
- Meta enters the crowded AI coding battle with Muse Spark 1.1(在新标签页中打开)
- Meta says its new AI model is ready to compete on coding(在新标签页中打开)
- Meta are apparently working on an open source variant of Muse Spark.(在新标签页中打开)
- Introducing Muse Spark 1.1(在新标签页中打开)
- Meta's Muse Spark 1.1 API pricing squeezes OpenAI and Anthropic as the AI price war heats up(在新标签页中打开)
- GPT-5.6 🚀, Muse Spark 1.1 ✨, ChatGPT Work 💼(在新标签页中打开)
- Meta removes controversial AI feature on Instagram after backlash(在新标签页中打开)
- Meta turns off the Instagram feature that let users make AI deepfakes of public accounts(在新标签页中打开)
中国开源浪潮. 中国模型现在占OpenRouter流量的30%以上;腾讯的HY3和GLM-5.2在家用硬件上运行,MiniMax计划发布2.7T参数的开源模型,DeepSeek的芯片设计表明了垂直整合,引发了美国的担忧。
- This is what Hy3 is capable of. Mother of god.(在新标签页中打开)
- llama.cpp: Hy3 PR + GGUFs(在新标签页中打开)
- Chinese AI models regularly pass 30 percent on OpenRouter as cost gap widens(在新标签页中打开)
- Chinese AI models are gaining ground with U.S. companies as OpenAI, Anthropic costs surge(在新标签页中打开)
- Why the rise of open source AI isn’t hurting Anthropic … yet(在新标签页中打开)
- Chinese AI startup MiniMax plans to open-source a 2.7 trillion parameter model later this year(在新标签页中打开)
- 4-bit GLM-5.2 (753B MoE) on 4× DGX Spark: 70.8% on Terminal-Bench 2.1 vs 81.0% for the full model(在新标签页中打开)
- Running GLM 5.2 on 4xGB10 with a 100G Switch, 330k ctx, ~25 t/s tg, ~650 t/s pp(在新标签页中打开)
- I created a 140 GB IQ2_XXS REAP quant of GLM 5.2 for coding. Looking for testers.(在新标签页中打开)
- Hy3 (295B MoE) and NVIDIA Nemotron-Labs-Audex-30B-A3B (audio-capable 30B MoE) GGUF quants(在新标签页中打开)
- Tencent-HY3 is the real deal on 128GB!(在新标签页中打开)
- The U.S. tech industry is increasingly anxious about the rising power and competitive price of open-source AI models from China — and whether the Trump administration will respond with yet another executive order | Politico(在新标签页中打开)
- China's DeepSeek developing its own AI chip, sources say(在新标签页中打开)
代理工程与基础设施
Bun被64个代理重写. Claude Fable 5编排了64个并行实例,在11天内将Bun从Zig重写为Rust,生成了超过一百万行代码,修复了128个错误,花费16.5万美元。
SWE-Bench Pro审计失败. OpenAI发现约30%的SWE-Bench Pro任务有缺陷,因此撤回了认可,呼应了Artificial Analysis早先的担忧,削弱了代理编程评估。
Modal融资3.55亿美元. Modal获得了3.55亿美元,用于构建专为AI代理设计的云基础设施,专注于为自主软件提供沙盒化、快速迭代的环境。
Cloudflare临时代理. Cloudflare推出了临时账户,允许AI代理在没有永久凭证的情况下部署Workers,如果无人认领则会过期,从而简化代理驱动的工作流程。
代理冲击本地GPU. 新的量化技术和经济高效的GPU设备现在可以在消费级硬件上运行MoE模型,但基准测试表明低位量化会严重降低代理任务的表现,敦促谨慎选择量化方案。
- Qwen3.5 122B is the best?(在新标签页中打开)
- Qwen3.6-27b does not understand software architechure.(在新标签页中打开)
- Qwen 3.6 Q2-FP8 Terminal Bench 2 and GPQA Scores(在新标签页中打开)
- Has anyone tested how quantization hits different capabilities separately? My results are surprising.(在新标签页中打开)
- 2.5x faster Qwen3.6 NVFP4 Unsloth quants(在新标签页中打开)
- Ultra budget 20GB vram with 448GB/s for $100 bucks.(在新标签页中打开)
- I benched quad 5060Tis for code generation with Qwen3.6-27B so you don't have to (it's really good)(在新标签页中打开)
安全、法律与可解释性
Jacobian Lens映射模型思维. Anthropic的Jacobian Lens揭示了一小组可言语化的表征,这些表征捕捉了模型的核心推理,实现了幻觉检测、输出引导,并且如社区演示所示,可以立即创建有害变体。
- Anthropic Research - "Verbalizable Representations Form a Global Workspace in Language Models"(在新标签页中打开)
- Claude's hidden inner monologue is now readable thanks to Anthropic's new Jacobian Lens(在新标签页中打开)
- I tested Anthropic’s new Jacobian Lens on open models, then it turned into a local-model hallucination router(在新标签页中打开)
- Anthropic found a hidden space where Claude puzzles over concepts(在新标签页中打开)
- I created a super harmful model ! :D (by tweaking it's J-Space!!!)(在新标签页中打开)
代理驱动的勒索软件. Sysdig记录了一个AI代理处理勒索软件攻击的技术执行过程,尽管操作是由人类设置的,这突显了当前代理恶意软件的能力和局限。
AI滥用危机. 一起诉讼指控xAI的Grok生成了CSAM图像,而博科圣地利用聊天机器人进行攻击策划;这些事件凸显了随着代理能力的扩展,迫切需要安全护栏。
苹果起诉OpenAI. 苹果提起联邦诉讼,指控OpenAI挖走400多名员工以窃取硬件商业机密,可能用于制造竞争性的AI智能手机,加剧了AI行业的法律斗争。
这是本周回顾 – 下周日见。