Data for AgentsHugging FaceIA
NVIDIA 发布 Nemotron 系列开放数据集，专注于 agent 训练所需的工具使用、多步推理、失败恢复等真实场景数据。
jeu · Quotidien tech · Numéro 3
2026-07-09
TL;DR du jour
共采集 30 条候选情报（LLM 综述不可用，以下为规则排序结果）。
À la une
3
China’s MiniMax Plans to Launch 2.7-Trillion Parameter Model
https://www.theinformation.com/briefings/exclusive-chinas-minimax-plans-launch-2-7-trillion-parameter-model According to The Information, MiniMax plans to launch a new-generation large language model with 2.7 trillion parameters. Sources revealed that the internal codename for this new model is M3 Pro. It is expected to be released and open-sourced as early as the third quarter of this year, with significant improvements in handling complex reasoning and multi-step tasks. This new model is much larger than MiniMax's current flagship model, M3 (428 billion parameters). Larger-scale artificial intelligence models are more capable of handling complex reasoning and multi-step instruction-based tasks.
Chaque matin, un digest tech fait pour vous
Le web montre la vue d’ensemble ; les abonnés reçoivent la leur — sélection IA selon vos intérêts, votre RSS privé intégré, avec les avis de la communauté, livrée chaque matin. Gratuit à vie.
44 numéros publiés · 150+ infos filtrées à 30 chaque jour
Actu IA
阿里团队提出 Agent 评测新范式 HSCodeComp，发现当前 Agent 准确率仅 45%，远低于人类专家 95%，揭示推理链过长与领域知识不足的结构性瓶颈。
TurnOPD: Making On-Policy Distillation Turn-Aware for Efficient Long-Horizon Agent TrainingarXiv cs.AIRecherche
arXiv:2607.05804v1 Announce Type: new Abstract: On-policy distillation (OPD) trains a student policy by matching a stronger teacher on the student's own trajectories, offering a promising framework for language agent training. However, its application to long-horizon agentic tasks remains insufficiently explored. We identify two key inefficiencies in vanilla agent OPD: (1) full-horizon rollouts often waste wall-clock resources on tail turns that provide weak and noisy KL supervision, and (2) tra
Hugging Face 集成 transformers 为 vLLM 的原生建模后端，让模型作者无需手动移植就能在 vLLM 中运行新架构。
OpenAI 分析指出 SWE-Bench Pro 等编码基准存在噪声问题，影响模型能力评估的可靠性，重新审视评测方法。
Dev & open source
Introducing GPT‑LiveSimon WillisonAutres
Introducing GPT‑Live OpenAI finally upgraded the model used by ChatGPT voice mode! I've had preview access for a few weeks in the iPhone app, and the new model is very impressive. It also has the ability to spin off harder tasks to GPT-5.5: For questions that require web search, deeper reasoning, or more complex work, it delegates to our latest frontier model behind the scenes and brings the result back into the conversation when it’s ready. While it works, GPT‑Live can keep talking with you and
Both vulnerabilities allow untrusted users to gain root privileges.
Android Bench is evolving, and developers can help guide that process.
Quoting Kenton VardaSimon WillisonAutres
Kenton Varda 禁止团队使用 AI 编写 PR 描述，认为 AI 只罗列代码细节而缺乏高层意图，适得其反。
Elon Musk's tech company released the newest version of Grok on Wednesday, promising a cheaper, more efficient alternative to other powerful AI models.
Échos de la communauté
评论区普遍对TypeScript 7的性能提升表示兴奋和祝贺，但也有人认为TypeScript仍有痛点，如配置复杂和与部分工具兼容性不佳。
评论区普遍认为Grok 4.5性价比高、速度快，性能接近Opus级别，但也有人质疑其实际表现和xAI的商业可持续性。
Cognition 发布 SWE-1.7 模型，在成本较低的条件下达到前沿级编程智能，验证 RL 后训练仍有巨大潜力。
微软发布 Flint，一种面向 AI agent 的可视化中间语言，支持 46 种图表类型，兼容 Vega-Lite、ECharts 等渲染库。
Mistral 推出 Robostral Navigate，一个 8B 参数机器人导航模型，仅用单 RGB 相机即可自主导航，性能超越多传感器方案。