🚀 Modèles & Releases
Anthropic has launched Claude Sonnet 5 (claude-sonnet-5), the next generation of its Sonnet model family, featuring a 1M token context window and 128k max output tokens. (https://docs.anthropic.com/en/release-notes/api#june-30-2026)
SpaceXAI (xAI) released Grok 4.5, described as an 'Opus-class model' with frontier performance on coding, knowledge work, and STEM. (https://techcrunch.com/2026/07/08/spacexai-releases-grok-4-5-which-elon-describes-as-an-opus-class-model/)
AionLabs released Aion-3.0-Mini, a multi-model roleplaying and storytelling system built on the DeepSeek family of models. (https://openrouter.ai/models/aion-labs/aion-3.0-mini)
⚙️ API & Dépréciations
Anthropic's Claude Sonnet 5 is available at an introductory pricing of $2 / $10 per MTok through August 31, 2026, with standard pricing of $3 / $15 thereafter. (https://docs.anthropic.com/en/release-notes/api#june-30-2026)
🔬 Recherche
A new paper, "FreqDepthKV: Frequency-Guided Depth Sharing for Robust KV Cache Compression in Long-Context LLM Inference," introduces an inference-time cache compression method to address the memory and bandwidth cost of KV caches. (https://arxiv.org/abs/2607.06519v1) Research on "Breaking Database Lock-in: Agentic Regeneration of High Performance Storage Readers for Database Bypass" tackles the bottleneck of data access guarded by database drivers like JDBC or ODBC. (https://arxiv.org/abs/2607.07696v1) A paper titled "Accurate, Interdisciplinary and Transparent Structure-property Understanding with Deep Native Structural Reasoning" explores structure-property relationships foundational to biology, chemistry, and materials science. (https://arxiv.org/abs/2607.07708v1) Other notable research includes "MANCE: Manifold Aware Concept Erasure" for removing target concepts from representations and "Do All Visual Tokens Matter Equally? Object-Evidence Preserving Token Merging for Vision-Language Retrieval" on token compression. (https://huggingface.co/papers/2607.03973, https://huggingface.co/papers/2607.04605)
🛠️ Outils & Produits
Google DeepMind announced an expansion of Managed Agents in the Gemini API, including features for background tasks and remote MCP. (https://blog.google/innovation-and-ai/technology/developers-tools/expanding-managed-agents-gemini-api/)
Ollama v0.31.2 was released, enabling flash attention on older NVIDIA GPUs (compute capability 6.x) and allowing iGPUs to offload vision models. (https://github.com/ollama/ollama/releases/tag/v0.31.2)
The Codex release v0.143.0 enables remote plugins by default and allows routing authentication and Responses API traffic through system proxies. (https://github.com/openai/codex/releases/tag/rust-v0.143.0)
Pydantic-AI v2.7.0 added support for Azure responses shorthand and the xAI grok-4.5 model. (https://github.com/pydantic/pydantic-ai/releases/tag/v2.7.0)
Llama.cpp commit b9911 introduced CUDA fusions for MMVQ post-scale operations. (https://github.com/ggml-org/llama.cpp/releases/tag/b9911)
Weaviate released v1.37.12 and v1.36.21, hardening async-replication, read-repair, and queue resilience, with no breaking changes. (https://github.com/weaviate/weaviate/releases/tag/v1.37.12, https://github.com/weaviate/weaviate/releases/tag/v1.36.21)
OpenHands released version 1.9.1, adding a super-admin management endpoint. (https://github.com/OpenHands/OpenHands/releases/tag/1.9.1)
New models on Hugging Face include NVIDIA's nvDock and Tencent's R3-rerank-0.6b and R3-embedding-0.6b. (https://huggingface.co/nvidia/nvDock, https://huggingface.co/tencent/R3-rerank-0.6b, https://huggingface.co/tencent/R3-embedding-0.6b)
🏢 Industrie & Business
AI startup Lovable is reportedly in talks to raise a $300 million round led by Menlo Ventures, which would double its valuation to $13.2B. (https://techcrunch.com/2026/07/08/lovable-reportedly-in-talks-to-double-its-valuation-to-13-2b/) A MIT Technology Review article discussed OpenAI CEO Sam Altman's promise regarding Americans sharing in the wealth AI creates. (https://www.technologyreview.com/2026/07/06/1140176/your-familys-300-stake-in-openai/) Another article outlined the foundational elements of AI architecture that IT leaders need to scale. (https://www.technologyreview.com/2026/07/07/1139413/the-foundational-elements-of-ai-architecture-that-it-leaders-need-to-scale/)
⚖️ Régulation & Société
Google's deepfake detector system was used to debunk an AI-generated fake image of Senator Mitch McConnell. (https://techcrunch.com/2026/07/08/googles-deepfake-detector-system-used-to-debunk-mcconnell-hoax-pic/) NYC public schools have put technology purchases on hold amid a heated debate over AI policy. (https://news.google.com/rss/articles/CBMixAFBVV95cUxQRW1DZFV3OVFpcm9CdWNNb0d4V2RITzV5dnVwMGJMMnFYb29BUjBSOGt0OWp0MkxTRExZdE93MmlYNHhMRklDSEJDal9WY2paYkpzWVV3ajlaNUlDdHcxSlNsVXFzMjNwQ3p0UWRzZDF5X3VEWjF4LUdWQmNpNXlpT0hyQnRzMDlxM2ZkNE85cVEyd201ZjhLM1djLVVYWXRlZ3BtZ2pWT2t4ak15RXBEQUdBRnhvWTlaUFBQZWdUYlpJQUZU?oc=5) A new generative AI policy and AI Hub website were offered by UA (University of Alabama). (https://news.google.com/rss/articles/CBMilAFBVV95cUxNa0ZvTDhVRmV5VTJRWGxrazBFdmR5bnhhbW1RS2VUcVNmMXo5OHRwaml6Q0t4cmotMGVzLUpEdVlsVlVnZHg2QXVLM0s3RUttSURFTU4xUm9CWUFyanNQZm1vUGtJRnUzd2t5cWhleGs0NTkydlE3RU9nVTViZnk1X0labVZHWkwtNmItbUFvQWwwTUxC?oc=5) An article explored how the First Amendment applies to AI regulation in hiring and health care. (https://news.google.com/rss/articles/CBMimAFBVV95cUxNU3FxVUxrbTc0VU5JZkk2UHRQUjZ2eVVLckd1UVlqT3YyOVVRZ2RVR3g2RUd2TFlMUW9hN2dyUWdWaXJxMDRIYmt1cVJPdXl4Q0QyQXhLcmNWZWVlUXNrNWxJckRwYzFnbmllRERyWFo5X1F0aFdWNmZ3VzcwcDB4RHFfaXpGNzIzTXFoSDZiVV9ka3U5TkFVQg?oc=5)
💡 À retenir
Anthropic's Claude Sonnet 5 launch sets a new price-performance benchmark with its 1M token context at introductory rates, directly impacting cost calculations for long-context applications until the end of August.