Brief

Researchers have developed ETCHR, a novel image editing model designed to enhance the visual reasoning capabilities of multimodal large language models (MLLMs). ETCHR decouples image editing from language understanding, employing a two-stage training process to improve how MLLMs interpret and manipulate visual information. This approach has demonstrated significant performance gains across various visual reasoning tasks when integrated with models like Qwen3-VL-8B, Gemini-3.1-Flash-Lite, and Kimi K2.5. AI

Researchers are developing new benchmarks to address the safety risks of AI agents, particularly in multi-agent and interactive environments. GT-HarmBench evaluates frontier models in game-theoretic scenarios, revealing significant failures in high-stakes situations. Boiling the Frog and AgentThreatBench focus on incremental attacks and indirect prompt injections that traditional benchmarks miss, assessing both task utility and security. These efforts aim to create more robust evaluations for AI systems operating beyond simple text generation. AI