🚀 Modèles & Releases
OpenAI released GPT-5.6 Luna Pro and GPT-5.6 Terra Pro, new reasoning-mode variants of their GPT-5.6 series models for handling complex tasks. (https://openrouter.ai/models/openai/gpt-5.6-luna-pro) New models on Hugging Face include NVIDIA's Nemotron-Labs-Audex-2B for audio-language tasks, NVIDIA's Kimi-K2.7-Code-DFlash, and Tencent's HiLS-Attention-7B for long-context text generation. (https://huggingface.co/nvidia/Nemotron-Labs-Audex-2B)
⚙️ API & Dépréciations
OpenAI is sunsetting its AI-powered browser, Atlas, after less than a year, moving some agentic browsing features to its desktop app and a Chrome extension. (https://techcrunch.com/2026/07/09/openai-is-shutting-down-atlas-but-its-ai-browser-ambitions-are-still-growing/) Meta removed a controversial AI feature on Instagram following user backlash, stating the feature "missed the mark." (https://techcrunch.com/2026/07/10/meta-removes-controversial-ai-feature-on-instagram-after-backlash/) The developer community is expressing concern over a potential discontinuation of Google's Gemini 2.5 Flash model. (https://discuss.ai.google.dev/t/please-dont-discontinue-gemini-2-5-flash/174246) Weaviate released v1.38.3 with fixes for async replication stability, BM25 search performance, and backup operations. (https://github.com/weaviate/weaviate/releases/tag/v1.38.3)
🔬 Recherche
Anthropic researchers developed a technique providing a clearer glimpse into the internal processes of large language models like Claude. (https://www.technologyreview.com/2026/07/09/1140293/anthropic-found-a-hidden-space-where-claude-puzzles-over-concepts/) New research benchmarks include UniClawBench for evaluating proactive agents on real-world tasks, OpenCoF for reasoning through video generation, and IdeaGene-Bench for scientific lineage reasoning. (https://arxiv.org/abs/2607.08768v1)
🏢 Industrie & Business
Apple has filed a lawsuit against OpenAI, alleging trade secret theft directed by senior leadership, including a former employee. (https://techcrunch.com/2026/07/10/apple-sues-openai-over-alleged-trade-secret-theft/)
🛠️ Outils & Produits
SGLang v0.5.15 was released, featuring production tuning for the GLM-5.2 NVFP4 model on Blackwell GPUs. (https://github.com/sgl-project/sglang/releases/tag/v0.5.15)
Llama.cpp update b9957 includes server improvements for tools and removes the apply_diff function. (https://github.com/ggml-org/llama.cpp/releases/tag/b9957)
The Llama Stack (ogx) released v1.2.0 with fixes for VertexAI logging and payload-bearing log limits. (https://github.com/ogx-ai/ogx/releases/tag/v1.2.0)
OpenAI's Codex released version 0.144.0, adding features for usage-limit credits, a new app-approval mode for writes, and interactive MCP tool authentication. (https://github.com/openai/codex/releases/tag/rust-v0.144.0)
Cline v4.0.8 added more models to its GCP Vertex provider and a free-form entry option for custom models. (https://github.com/cline/cline/releases/tag/v4.0.8)
OpenHands released cloud version 1.46.0 and version 1.11.0, with updates to archive manifests, BYOR key configuration, and agent profiles. (https://github.com/OpenHands/OpenHands/releases/tag/cloud-1.46.0)
⚖️ Régulation & Société
U.S. lawmakers discussed online child safety and AI policy at a bipartisan summit. (https://news.google.com/rss/articles/CBMivgFBVV95cUxQQlFjcGYxa1lfX3JMaFo3ZXpoRzlrQjVod014RGw2dVFKcVRwVUNGQVNLME9IYlJsaElLbUJuUDRTSW5EWXdJZGs5OHY4VkZhVHRLekp6dWZEQU91XzByVVJXTU9kSFJQVGsteWV1NWxhWjlBU2QxTkw4RG5ycS1BQWVJNjNPemdyRXJ1SWVGMVByTDlnWlB2YnJqOXNQWDk1T1IyS3ExRmdKV2RUcjlLRWVIZk5iU3R2d1VtLWtn?oc=5) Opinion pieces debated the constitutionality of private AI regulation and noted that regulators are seeking concrete evidence over broad policy. (https://news.google.com/rss/articles/CBMiggFBVV95cUxNRDRaT2MxckhuMmNWVzJXZG1tcV9VbDFQdGZUTGNmb1Q0WmpQaEtxUklPZmwzbTRlNDY5QVNaS3E5N1ZVSXdiYmVPeXFScEM1aFRMcFNCVGxDUmU5VUM1MkJrOGJ3aXl0d0tzMUJLMnJjM0ZBR1ZYWkNOLVZBaGdPaW53?oc=5)
💡 À retenir
The most significant action is OpenAI's shutdown of its Atlas browser, signaling a strategic pivot away from a standalone browsing product less than a year after launch, while Meta's removal of an Instagram AI feature highlights the ongoing user and regulatory sensitivity around AI deployment in consumer apps.