Daily LLM API Advisory — June 24, 2026

ApiDelta · 2026-06-24 · 276 words · apidelta.maxiaworld.app

🚨 Breaking No breaking changes reported.

🗑️ Dépréciations No deprecations reported.

💰 Pricing No pricing changes reported.

🆕 Nouveautés - Mistral OCR 4 (mistral-ocr-4-0) is now live; mistral-ocr-latest points to it. New include_blocks parameter returns paragraph-level bounding boxes with structural labels (text, title, list, table, etc.). Changelog - Anthropic released Claude Fable 5 (claude-fable-5) and Claude Mythos 5 (claude-mythos-5) with 1M token context, 128k max output, and adaptive thinking. Release notes - TensorRT-LLM v1.3.0rc19 adds support for NVIDIA Wan2.2-T2V quantized checkpoints, T5/BART and MiniMax-M3 models in the PyTorch backend. Release - llama.cpp builds b9776, b9775, b9767, b9763, b9761 bring Vulkan FA overflow fix, draft context error check, WebGPU MTP optimization, tool call IDs in API, and server-side model download progress. b9776 and others

🌐 Actualité IA No significant industry signals outside provider updates.

💡 Conseil du jour Update your Mistral OCR integration to point to mistral-ocr-latest and enable include_blocks to get structured paragraph-level output with labels. This can simplify downstream extraction pipelines.

#api#llm#en