🚀 Modèles & Releases
Anthropic's Dreams research preview now supports Claude Fable 5 and Claude Sonnet 5 models. (https://docs.anthropic.com/en/release-notes/api#july-10-2026)
Kwaipilot released KAT-Coder-Air V2.5 and KAT-Coder-Pro V2.5 models. (https://openrouter.ai/models/kwaipilot/kat-coder-air-v2.5, https://openrouter.ai/models/kwaipilot/kat-coder-pro-v2.5)
OpenAI's GPT-5.6 Luna Pro is now available, serving the same underlying model as GPT-5.6 Luna with reasoning.mode set to pro for complex tasks. (https://openrouter.ai/models/openai/gpt-5.6-luna-pro)
CohereLabs released cohere-transcribe-arabic-07-2026, a model for Arabic speech recognition. (https://huggingface.co/CohereLabs/cohere-transcribe-arabic-07-2026)
🔬 Recherche
Anthropic published a significant but nuanced AI discovery, analyzed by MIT Tech Review. (https://www.technologyreview.com/2026/07/13/1140343/what-anthropics-latest-ai-discovery-does-and-doesnt-show/) A new arXiv paper introduces "Requential Coding," a method pushing the limits of model compression using self-generated training data. (https://arxiv.org/abs/2607.11883v1) Another arXiv paper reviews the foundations, progress, and opportunities of metacognition in LLMs. (https://arxiv.org/abs/2607.11881v1) A theoretical framework explaining the emergence of inductive reasoning abilities in Transformer language models was published on arXiv. (https://arxiv.org/abs/2607.11875v1) ABot-AgentOS, a general robotic Agent Operating System with lifelong multi-modal memory, was detailed in a new paper. (https://huggingface.co/papers/2607.10350) ABot-N1 is a new foundation model aimed at unifying reasoning for visual language navigation tasks. (https://huggingface.co/papers/2607.10383) LightMem-Ego proposes a lightweight multimodal memory system for personal AI assistants on mobile and wearable devices. (https://huggingface.co/papers/2607.11487)
🏢 Industrie & Business
Video-generation startup PixVerse raised $439 million, with its valuation soaring past $2 billion. (https://techcrunch.com/2026/07/13/video-generation-startup-pixverse-raises-439m-valuation-soars-past-2b/) A TechCrunch article discusses why established, successful tech figures are intensely focusing on AI again, driven by FOMO and profit potential. (https://techcrunch.com/2026/07/13/already-rich-already-successful-why-the-last-wave-of-tech-winners-is-grinding-again/) Uber's Chief Product Officer discussed the company's financial-services ambitions, its relationship with Waymo, and how AI is becoming noticeable for riders and drivers. (https://techcrunch.com/2026/07/13/ubers-product-chief-on-hotels-robotaxis-and-why-the-company-doesnt-want-to-be-everything-for-everyone/)
🛠️ Outils & Produits
llama.cpp released version b9992, featuring a refactor of the CUDA MMQ kernel configuration and fixes for Blackwell config. (https://github.com/ggml-org/llama.cpp/releases/tag/b9992)
Camel AI released v0.2.91a5, including a fix to avoid re-entering an agent in screenshot analysis and the first phase of a semantic cache. (https://github.com/camel-ai/camel/releases/tag/v0.2.91a5)
Goose released v1.42.0, adding an OllamaCloudProvider with dynamic model discovery, per-message usage stats UI, and cost tracking. (https://github.com/aaif-goose/goose/releases/tag/v1.42.0)
Pydantic AI released v2.9.1, fixing bugs in JsonSchemaTransformer recursion and timeout handling for MistralModel. (https://github.com/pydantic/pydantic-ai/releases/tag/v2.9.1)
Ollama released v0.32.0, launching a new interactive agent experience that runs when you type ollama and renaming its Codex App integration to ChatGPT. (https://github.com/ollama/ollama/releases/tag/v0.32.0)
OpenAI's Codex released version 0.145.0-alpha.9. (https://github.com/openai/codex/releases/tag/rust-v0.145.0-alpha.9)
Cline CLI released v3.0.40, adding a manual API key escape hatch and fixing issues with provider config reloading and auto-updates. (https://github.com/cline/cline/releases/tag/cli-v3.0.40)
⚖️ Régulation & Société
An opinion piece from the Lowy Institute argues that current AI regulation is answering the wrong question. (https://news.google.com/rss/articles/CBMijwFBVV95cUxQTlVnLVhoZWN3VFpiSHdGT3VMQ0liLVV3TnFxX2V2Zk13WjFwVjJmQ2kyczlHMnM1NF9LSW8zRFJxT21fZlJ3WkRDSGZIbVBDWkxmRmNjcUwwUXpYT3R0TEVGUWNoYjltQWtZX0RtM3ctTngxNjlLSkVib2Nhc0NzbWtadGhEeTZoQ2pvTGRqaw?oc=5) Young U.S. lawmakers discussed online child safety and AI policy at a bipartisan summit. (https://news.google.com/rss/articles/CBMixgFBVV95cUxOSzBKWUpueFU1U1JoU2YxX1QzZU9FWmUxbXdnSzJ3MHdtRTU3SW1BbFJCQjF3eGcwNlRwRm8ySkZOZzREcUZhTUp3MkwtX0NDUkdublROMDM2RWR0RnZJdUVZbTVpNTdmUXFLaGJCa0hsdnJxbi1WVEwxZWZOLUVyWVlWUEZPYTRZQTR3TkJBRHZkZmZFeVQ0MTFtVm1jS0V4YXNhWUFYQjZwMF9Pd1BKRExpaU5OcjRHWFpUTWpZSXplZkREZWc?oc=5) An article highlights the widening AI policy gap in higher education, extending beyond academic integrity concerns. (https://news.google.com/rss/articles/CBMitwFBVV95cUxNbHhxTXNNWDJhUGJ3MG5KU2ZmM1FzenQyaG16djBSVmN2elNINEw4TVo5ZlZkOXJFTlhlQWQxakZUZmppb25PVUNEYVVxQlR6WnFRYnIzNkQyQmhVLWlBcnliRURPUXluVnJNeFdVaF9tcm5ZQTlnS3N0d3J4a1JEbVN6SkRneE5QMnR1cXhobnBvOVBSQkNENDRTV0dGZ3ZtTDlSQ1pCTTRHa3k5bVI3TFF3RS0zTUU?oc=5) Samsung Health app reportedly threatened data deletion if users opt out of AI training, sparking discussion. (https://neow.in/cWsyMTV3) A user-submitted article argues that "AI is a bad tool," generating debate. (https://bytecode.news/posts/2026/07/user-submission-ai-is-a-bad-tool) An article ponders the future of work, asking "What will be left for us to work on?" (https://www.normaltech.ai/p/what-will-be-left-for-us-to-work)
💡 À retenir
PixVerse's massive $439M funding round signals intense investor confidence in video-generation AI, pushing its valuation over $2B as it aims to expand its world model offering globally.