2026年7月8日
AI 消息
这周值得知道的 AI 消息。
OpenAI ships gpt-realtime-2.1 for faster, cheaper voice agents
OpenAI added gpt-realtime-2.1 and a cheaper mini version to its API for low-latency voice agents. They handle noise, silence, and interruptions better, and response latency dropped by at least 25%.
看原文 →NVIDIA releases Audex: one open model for speech, translation, and TTS
NVIDIA released Audex, a 30B open model that handles speech recognition, translation, text-to-speech, and audio generation all in one model, while keeping the text intelligence of its backbone nearly intact.
看原文 →Cohere open-sources Transcribe Arabic, built for Arabic's hardest speech
Cohere open-sourced Transcribe Arabic, a 2B speech recognition model built for Arabic dialects, Arabic-English code-switching, and tricky vocabulary. It beats Whisper and Meta's OmniASR on Arabic benchmarks, under an Apache 2.0 license.
看原文 →Meta just launched a new AI generator, Muse Image, and users are already pushing back over use of their photos
Meta launched Muse Image, a new AI image generator, and users are pushing back over a feature that lets people create AI images from other Instagram users' public photos unless they opt out.
看原文 →Ant Group's Robbyant Open-Sources LingBot-Vision: A 1B Boundary-Centric Vision Foundation Model for Dense Spatial Perception
Ant Group's Robbyant open-sourced LingBot-Vision, a 1 billion parameter vision model that uses boundary modeling to improve dense spatial perception. This model outperforms larger models in certain tasks and will be used to initialize LingBot-Depth 2.0.
看原文 →From Hugging Face to Amazon SageMaker Studio in one click
Hugging Face and AWS added a one-click path from any model page straight into Amazon SageMaker Studio, with the model pre-loaded and the environment already set up, so you can go from finding a model to trying it in minutes.
看原文 →Curry, bagels … and AI? Londoners fight plan for huge datacentre in Brick Lane
Residents and a London council oppose plans for a new datacentre in Brick Lane, arguing affordable housing needs are more pressing than AI-driven trading.
看原文 →Data centers' energy demand threatens Trump's "Made in America" plan
Rising energy demands from data centers could jeopardize US President Trump's "Made in America" manufacturing initiative due to increased pressure on electricity prices in industrial regions.
看原文 →Avoid AI atrophy - new tool promises to reverse vibe coding skills decay
A new command-line tool called Atrophy CLI aims to help developers maintain and improve their coding abilities by providing regular practice challenges.
看原文 →OpenAI (@OpenAI)
GPT-5.6 Sol, along with Terra and Luna, will launch publicly this Thursday. We're expanding preview access globally now.
看原文 →Claude (@claudeai)
Claude Cowork is coming to mobile and web. Hand Claude a task at your desk and pick up the finished work from your phone. Close the laptop and Claude keeps going. Beta is rolling out over the next several weeks starting with the Max plan, with more plans to follow.
看原文 →ClaudeDevs (@ClaudeDevs)
6 months of Claude Max 20x, on us. We're expanding Claude for Open Source to more of the community. If you're a maintainer, a core contributor, someone landing PRs across the ecosystem, or someone keeping a critical package alive, apply today!
看原文 →配音与开源
现在值得一看的开源语音、配音工具。
MisoLabsAI/MisoTTS
MisoLabsAI released MisoTTS, an 8 billion parameter text-to-speech model designed to generate expressive and emotionally nuanced speech.
看原文 →elder-plinius/T3MP3ST
GitHub user elder-plinius released T3MP3ST, an autonomous red team platform using a multi-agent system for offensive security testing.
看原文 →synthetic-sciences/openscience
Synthetic Sciences released openscience, an open-source AI workbench built to speed up scientific research, giving researchers ready-to-use tools to put AI to work in their studies.
看原文 →用邮件接收下一期
8 种语言任选，随时可以退订。