模型发布/更新
-
NVIDIA Alpamayo 2 Super opens for commercial use, a frontier open-source model for Robotaxi and autonomous driving
NVIDIA Alpamayo 2 Super is now open for commercial use. Built on Cosmos 3 Super Reasoner and post-trained with…
-
SenseTime SenseNova U1 goes open-source: unified reasoning and image generation
SenseTime released the open-source model SenseNova U1, which can perform reasoning and image generation simultaneously…
-
OpenRouter adds FLUX 3 Video unified multimodal model
@bfl_ai's FLUX 3 Video is now available to everyone on OpenRouter. A unified multimodal model family for video, audio,…
-
蚂蚁百灵发布Ling-3.0-flash开源权重
今天,我们发布了 Ling-3.0-flash 的开源权重。🎉 官方 BF16 和 FP8 量化版本现已可用,您可以根据自己的硬件、性能要求和部署需求选择最合适的版本。
-
Tencent Hunyuan releases Hy ASR 3.0 preview: speech recognition that truly understands context
Tencent Hunyuan released the next-generation speech recognition model Hy ASR 3.0 preview, based on the large language…
-
Black Forest Labs releases FLUX 3 Video with up to 20-second HD video and native audio
Black Forest Labs announced that FLUX 3 Video generation capabilities are now generally available through the BFL API…
产品发布/更新
-
Reflex open-sources XY: a Rust-based ultra-fast Python plotting library that keeps 100 million-point charts interactive
Reflex AI released XY, an Apache-2.0-licensed Python interactive 2D plotting library. Through a Rust native core,…
-
ModelBest open-sources ForgeStencil: automatically optimizing 100+ industrial and scientific software programs per week with zero human intervention
ModelBest, together with OpenBMB, open-sourced ForgeStencil, the world's first AI optimization system supporting…
-
OpenRouter launches ori CLI with out-of-the-box optimized configs for Claude Code and other harnesses
OpenRouter released ori CLI. After installation and login, users get optimized configurations for harnesses such as…
-
SpecForge v0.3.0 released with unified decoupled and colocated speculative decoding stack, open SpecBundle draft models
SpecForge v0.3.0 separates target model inference from draft model training, supports speculative decoding algorithms…
-
Cloudflare 让智能体通过本地追踪调试 Workers
Cloudflare 即日起在 wrangler dev 和 vite dev 中自动为本地 Worker 调用捕获 OpenTelemetry 追踪,无需安装 SDK 或配置智能体。智能体可通过 Local Explorer API 查…
行业动态
-
Anthropic signs $10 billion compute deal with months-old cloud startup Volta
Anthropic signed a $10 billion compute deal with Volta, a cloud startup founded only months ago, equivalent to about…
-
MIIT releases China's first mandatory national standard for L3/L4 autonomous driving system safety, effective July 2027
The mandatory national standard "Intelligent Connected Vehicles—Safety Requirements for Automated Driving Systems" (GB…
-
GPT-5.6 Luna price cut of 80% is permanent
Some people clearly misunderstood, but the 80% price cut for GPT-5.6 Luna is not a temporary gimmick—it is permanent.…
-
Mariano-Florentino (Tino) Cuéllar 加入 Anthropic 出任首席全球事务官
Anthropic 任命 Mariano-Florentino (Tino) Cuéllar 为首任首席全球事务官,负责全球政策、国际战略参与及政府关系事务。Cuéllar 曾任卡内基国际和平基金会主席、加州最高法院法官,并自 2026…
-
OpenAI 说明第三方网络安全评估事件并公布新保障措施
OpenAI 就近期第三方网络安全评估事件作出说明,并公布新的保障措施以强化 AI 模型测试与评估流程。相关措施旨在提升模型评估的安全性与可靠性,确保第三方测试在受控环境下进行。
-
AI 领袖提出 SAFE 指南,强化智能体网络安全透明度
Linux 基金会发布共享 AI 事件交换(SAFE)指南征求意见稿,旨在将智能体网络安全事件转化为全生态共享防护。
-
Google 发布 Gemini 3.6 Flash 等三款新模型及 Gemini Robotics ER 2
Google 在 7 月推出三款新 Gemini 模型——Gemini 3.6 Flash、3.5 Flash-Lite 和 3.5 Flash Cyber,面向生产级 AI 智能体,主打更高 token 效率、更低延迟和更可靠性能。
技巧与观点
-
MiniMax-H3 runs on Apple Silicon via MLX port
MiniMax released MiniMax-H3, a general omni-modal generation system that accepts text, images, audio and video and…
-
Building an advanced AI skill security audit pipeline with NVIDIA SkillSpector, LangGraph, YARA rules, SARIF and CI policy gates
This tutorial demonstrates how to use NVIDIA SkillSpector to assess the security posture of AI skills, build a…
-
Cloudflare 如何用软件工厂将 Astro 的 GitHub issue 数降至零
Cloudflare 在 Astro 仓库上运行自动化 triage 流水线,通过隔离的 AI 子代理复现、诊断并修复 bug,将开放 issue 从 200 多个降至约 30 个,预计下月归零。该流水线由 issue 标签驱动,修复后自…
-
How GitHub uses stacked pull requests to break down giant AI-generated code
GitHub described how stacked pull requests solve the difficulty of reviewing giant code generated by AI coding agents.…
-
如何用 LangSmith 评估语音智能体
LangChain 官方博客介绍如何用 LangSmith 评估语音智能体,覆盖执行、结果与来电者体验三个层面。评估手段包括 LangSmith traces、代码评估器、LLM judges 和人工审查,帮助开发者系统化验证语音智能体的…
-
From zero, learn to build your first hardware with Codex
Starting from zero experience, the author spent five days building a cat-paw hardware device that reminds users to…
-
Can the Jevons paradox last: how AI pricing tiers support growing compute demand
Tech executives widely say compute supply still cannot meet demand, yet AI pricing is rising: Anthropic's Fable 5,…