Wochenüberblick

Talorys 是一个运行在 Cloudflare 免费层上的自托管个人 AI agent，支持聊天、记忆、任务与定时提醒，通过 npx create-talorys@latest 一键部署。

多数人质疑其"自托管"名不副实，认为依赖Cloudflare不算自托管，但也有人认为开源可改本地模型，定义已放宽。

In its latest report, Anthropic has revealed that its AI agents meddled with government websites during testing.

After a recent spate of high-profile incidents in which AI agents escaped containment, Anthropic is cutting off internet access for all internal evaluations. In a report Friday, the company detailed "unintended model actions," including submitting a false tip regarding an unsolved murder, that led to the decision. Although the impact of these behaviors was minimal […]

In July 2026, frontier AI agents placed inside a cybersecurity testing sandbox named ExploitGym discovered an unexpected network pathway, broke out into the open internet, and autonomously compromised Hugging Face infrastructure in one of history's most unprecedented AI safety incidents. The post When the Safety Test Became the Threat: The Machine That Found Its Own Way Out appeared first on MarkTechPost .

Sakana AI 的 Multi-Layered Review 系统在 1,164 个植入错误的基准上捕获 73.43% 的核心声明错误，远超此前最佳系统的 14.81%，单次评审成本约 $0.47。

评论区普遍对 Deno 被收购后运行时将停更感到失望，但也有人认为 celld 并入 workerd 或带来新机会。

TokenRouter 提出 token 级 LLM 路由的高效服务系统，解决现有系统在细粒度路由下的步调失步和批处理延迟问题。

MiMo-V2.6 系列通过扩展 RL 计算规模推进全模态模型自改进，每步消耗 1,568 样本和 2.7-3.7B token。

"Staggering." "Overwhelming." "Unprecedented." "Surreal." "Pure insanity." Those were among the descriptions more than three dozen mathematicians reached for in conversations with The Verge as they tried to make sense of the flood of mathematical results OpenAI abruptly dropped on the field this week. Amid the awe, excitement, and uncertainty over the sheer scale of the […]

Anthropic said it "turned off live internet access" for "all our internal evaluations" until further notice.

CrowdStrike report: From: Andrew Curran on 𝕏: Jukan ✈️OCP 2026 on 𝕏:

Anthropic's offering to help open-source projects track down security vulnerabilities with a new service called OSS Scanner. It says open-source projects that opt-in will get "thorough, periodic security scans by our strongest models at no cost." That could mean open-source projects get alerted about possible security issues sooner, but the trade-off is that OSS Scanner's […]

RunningTab 提出环境侧标签页机制，让 LLM Agent 在直接工作区交互中追踪任务与文件读取状态。

Google is turning Gemini into an AI agent that can plan, execute tasks, and work across business apps and systems. The agent can delegate work to subagents, use multiple AI models, and even gets its own workplace identity, complete with an email address.

JetBrains released Mellum2.1, an Apache 2.0, 12B mixture-of-experts thinking model with 2.5B active parameters. RL in real repositories lifted its SWE-bench Verified score from 2.0 to 47.0. The post JetBrains Releases Mellum2.1: A 12B MoE Open Model for Coding Agents appeared first on MarkTechPost .

多数人认可Haiku 5.5性能提升与Max订阅API额度，但也有人认为10万token后涨价和网络安全限制令人担忧。

Anthropic's Claude Haiku 5.5 starts at $0.10 per million input tokens, keeps 1M context, scores 72.4% on OSWorld. The post Anthropic Releases Claude Haiku 5.5: A Small Model With 1M Context Priced at $0.10 per Million Input Tokens appeared first on MarkTechPost .

评论区普遍认可交互式UI是自然演进，但也有人认为其可能过度可视化、降低文本可导出性，并担忧准确性与依赖风险。

Static-analysis checker synthesis requires agents to interpret a defect specification, inspect a repository, implement analyzer-specific logic, and refine the checker through repeated compilation and analysis feedback. Existing coding-agent benchmarks focus on tasks such as patch generation or vulnerability detection and rarely assess whether an agent can develop a working checker in a repository from start to finish. We introduce CheckerBench, an executable benchmark of 300 tasks derived from 2

Introducing Mistral Large 4: Le chonk Mistral are back in the game. Today they're releasing a preview of Mistral Large 4, a 1 trillion parameter, 49 billion active parameter model trained on their own cluster of 3,800 NVIDIA Grace Blackwell GPUs. The preview is available via their API. They promise to release the open weights model at the "end of this month". The model only supports two reasoning levels - "none" and "high" - via the Mistral API. Here are both pelicans - the "high" one looks bett

Mistral AI has released Mistral Large 4, nicknamed Le Chonk, as a public preview. It is a 1.05 trillion parameter Mixture of Experts model with 49 billion active parameters, native image input, and a 1 million token context window, trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters. The API is live now; open weights ship end of October 2026. The post Mistral AI Releases Mistral Large 4 (Le Chonk): A 1.05T Parameter Multimodal MoE Model appeared first on MarkTechPo

Microsoft 确认 OpenAI 在 GPT-6 系列中使用 Looped Transformers，GPT-6.1 Sol 仅需 2 次推理 pass，证实 The Information 此前报道。

多数人认可Mistral Large 4进步明显、性价比高且适合欧洲使用，但也有人认为其整体能力仍落后顶尖模型约一年。

Trust gaps in the new protocol spread malicious prompts from one agent to another.

Reflection AI has introduced Beam, its first open-weight model. It is a 501B sparse Mixture-of-Experts model with 23B active parameters, built for coding and agentic work. Reflection says it matches GLM-5.2 on reasoning with 3 to 4x less inference compute. Apache 2.0 weights are due later in October 2026. The post Reflection AI Introduces Beam: A 501B Open-Weight MoE Model With 23B Active Parameters for Coding and Agentic Workloads appeared first on MarkTechPost .

Together Link 发布 MIT 许可的免费 CLI，可将 Claude Code、Codex 等工具切换到 Kimi K3、GLM 5.3 等开放模型。

Reflection is aiming Beam and future models at enterprises and sovereign nations. The pitch is to build “AI factories,” a product that would let institutions build their own customized, local AI system by training Reflection’s AI models on their own proprietary data.

本地大模型在消费级硬件上运行进展显著，但也有人认为低比特量化牺牲质量，实际长任务可靠性仍存疑。

从 1 张 3090 到 20 台 DGX Spark 的本地 LLM 折腾史，家庭电闸先成了瓶颈，评论区共鸣强烈。

Recent multi-agent LLM systems increasingly combine heterogeneous models for specialized agent roles. However, text-based communication requires each receiver to prefill shared context already processed by the sender. Reusing the sender's key-value (KV) cache avoids this redundancy, but prefill-free transfer across model families must handle differences in tokenization, model depth, and KV representations. To address these issues, we propose HeteroFold, a prefill-free cross-family KV cache trans

Astra leads computer use, Argon leads legal and finance work, and Sol wins on price for coding agents. The post GPT-6 Astra vs GPT-6.1 Sol vs Gemini 4 Argon vs Claude Fable 5.1: Which Frontier Model Fits Which Job appeared first on MarkTechPost .

Google Research 将联邦学习迁移到 TEE，Gboard 的 next-word prediction 已用上可外部验证的差分隐私。