13 articles — 🔴 3 critical , 🟡 5 important , 🟢 5 interesting
Claude Opus 4.8 is Anthropic's new most capable generally available model, launched on May 28, 2026. It brings a one-million (1M) token context window by default on the Claude API, Amazon Bedrock and Vertex AI, 128k maximum output tokens, and adaptive thinking that triggers reasoning only when needed.
Qwen-VLA is a unified embodied foundation model from the Qwen team that integrates vision, language and action for diverse robotic tasks such as manipulation and navigation, across different robot platforms. The paper, with 40 authors including Junyang Lin and Jingren Zhou, achieves 97.9% on the LIBERO benchmark and generalizes strongly to new environments and embodiments.
On May 28, 2026, Google declared general availability (GA) of two native visual models: gemini-3.1-flash-image (Nano Banana 2) and gemini-3-pro-image (Nano Banana Pro). The novelty is video-to-image generation in the flash variant, which generates thumbnails, movie posters and infographics from video files or public YouTube URLs.
FLUX VTO is a new model from Black Forest Labs for virtual clothing try-on. It lets users see themselves in garments before purchase, with high accuracy and consistency of garment rendering, and generation time under four seconds.
On May 28, 2026, vLLM introduced standardized APIs for weight synchronization and improved support for asynchronous reinforcement learning. It addresses the fragmentation in which every RL framework ad-hoc implements weight transfer between training and inference, and fixes stability problems in large distributed deployments.
Data Formulator 0.7 is an open-source system from Microsoft Research that connects fragmented data sources in enterprises and supports iterative analytical workflows. It brings persistent data connectors and context-aware AI agents, enabling analytics without knowledge of SQL or programming.
Mistral Search Toolkit is an open-source framework for building production search pipelines for AI applications. It unifies ingestion, retrieval and evaluation under a single interface and can be deployed in the cloud, on-premises or at the network edge.
OpenJarvis v1.0 is an open-source framework for building personal AI agents that run locally on your own hardware, with built-in Ollama integration. It is a local-first approach in which models stay local and the cloud is optional, with tracking of energy consumption, cost and latency.
Vibe is Mistral's new unified agent, unveiled on May 28, 2026 on the chat.mistral.ai platform. It launches with two primary modes — Work for long-horizon productivity and Code for software development — while retaining the legacy Chat experience, and unifies Skills, Workflows, Connectors and Libraries.
LongDS-Bench is a new benchmark for evaluating AI agents on multi-turn data analysis tasks. It contains 68 tasks and 2,225 turns from real Kaggle notebooks, and five leading models show significant degradation over long horizons — the best achieves only 48.45% average accuracy.
