🚨 Breaking
- Weaviate v1.38.2 and v1.37.10 released (2026-06-25 and 2026-06-24). Although changelogs state no breaking changes, these are critical patches fixing async replication, tenant cap enforcement, goroutine leaks, and more. Immediate upgrade recommended to avoid integration breakdowns. v1.38.2 | v1.37.10
🗑️ Dépréciations
Aucun changement.
💰 Pricing
Aucun changement.
🆕 Nouveautés
- Weaviate introduces generative-deepseek module in v1.38.2. New module for DeepSeek integration.
- llama.cpp multiple updates: b9788 (SYCL tensor parallelism), b9780 (Vulkan build failure detection), b9789 (MoE quant fix), b9777 (LFM2.5-ColBERT-350M and LFM2.5-Embedding-350M models), b9776 (Vulkan bias before softmax), b9775 (draft context error check).
- TensorRT-LLM v1.3.0rc19: Support for NVIDIA Wan2.2-T2V quantized checkpoints, T5, BART, MiniMax-M3 in PyTorch backend, and more.
- Sakana unveils Fugu Ultra, a learned multi-agent orchestration system with improved performance. Available via OpenRouter. Fugu Ultra
🌐 Actualité IA
Aucun signal.
💡 Conseil du jour
Upgrade Weaviate to v1.38.2 (or v1.37.10 if on 1.37.x) today. These releases include fixes for replication, tenant limits, and S3 backup goroutine leaks that could silently degrade production performance. Validate your generative-deepseek module configuration if you plan to use DeepSeek models.