Brief

A new open-source index called AgentTape ranks AI models based on a blend of benchmark performance, actual usage, cost, and speed. Currently, OpenAI's GPT-5 models dominate the top rankings, with GPT-5.5 specifically excelling in quality benchmarks but lagging in adoption due to its newness and price. The index aims to provide a more holistic view of model performance beyond theoretical benchmarks, reflecting real-world utility. AI

A new benchmark called ESI-Bench has been released by Fei-Fei Li's team to evaluate embodied spatial intelligence in AI. Unlike previous benchmarks that assumed optimal observation, ESI-Bench requires AI agents to actively take actions to gather information, closing the perception-action loop. Initial tests with leading models like GPT-5 and Gemini revealed that current AI struggles with active exploration and decision-making, exhibiting "action blindness" and metacognitive deficits, indicating that the primary challenge lies in strategic action rather than pure perception. AI

A recent benchmark evaluated six large language models on their ability to extract structured data, specifically JSON, from customer support emails. The analysis found that Anthropic's Claude Haiku 4.5 offered the best value, achieving high accuracy at a significantly lower cost compared to more powerful models. While Gemini 2.5 Flash was fast and inexpensive, it struggled with accuracy, particularly in hallucinating data. The study suggests using Haiku for most extraction tasks, Sonnet for more complex reasoning, and avoiding more expensive frontier models for simple data extraction. AI

A new benchmark called ARFBench reveals that human engineers still significantly outperform AI models like GPT-5 and Gemini in diagnosing system failures. The results challenge the marketing claims of AI's full autonomy in production environments, highlighting the current limitations of AI in complex troubleshooting tasks. AI

Several recent posts explore advancements and applications in AI agents, particularly for coding and reasoning tasks. Topics include building autonomous coding agents that can open GitHub pull requests, using patterns like Continual Harness for self-improving agents, and integrating tools like Cursor into agent workflows. The limitations of LLM reasoning in causal inference and new approaches to browser fingerprinting for web scraping are also discussed, alongside efforts to automate hyperparameter tuning for machine learning models. AI

Microsoft has introduced four new AI-related certifications to address the growing demand for AI professionals. Separately, there are reports that Elon Musk's xAI may have failed to pay a $420 fee for tax data. Additionally, Microsoft's M365 Copilot, initially criticized, is reportedly seeing a turnaround, with GPT-5 cited as a potential catalyst for improved performance and a shift away from prompt engineering expertise. AI

Researchers from Shanghai Jiao Tong University and the Qwen team have introduced CodePercept, a novel approach to enhance large language models' visual perception capabilities, particularly for STEM tasks. Their research suggests that improving visual perception, rather than just reasoning, is the key bottleneck for models tackling science and math problems. CodePercept leverages code as a precise language for visual understanding, enabling models to generate executable code that accurately represents image content, thereby overcoming the inherent ambiguity of natural language descriptions. AI

Researchers have developed a method for language models to predict the success of scientific research ideas before experimentation. By training models on a dataset of comparative idea evaluations, they achieved significant accuracy in forecasting empirical outcomes. This approach, particularly when framed as a reasoning task using Reinforcement Learning with Verifiable Rewards, allows even smaller, compute-efficient models to act as objective verifiers, potentially accelerating autonomous scientific discovery. AI

Researchers have developed DrugRAG, a novel retrieval-augmented generation pipeline designed to enhance the performance of large language models (LLMs) on pharmacy-related question-answering tasks. In their study, they evaluated ten LLMs, finding that GPT-5 and o3 performed best on a 141-question dataset. DrugRAG, which integrates structured drug information without altering model architecture, significantly improved accuracy across several models, particularly smaller open-source ones, by up to 21 percentage points. AI

Researchers have introduced BEAR, a new benchmark designed to evaluate and diagnose the skill-level capabilities of embodied multimodal large language models (MLLMs). This benchmark decomposes embodied tasks into 14 distinct atomic skills, providing more granular insights into model failures than previous task-level evaluations. Evaluations on BEAR revealed that perceptual limitations and unstable spatiotemporal modeling are significant bottlenecks for current MLLMs. To address these issues, the team developed BEAR-Agent, a conversational agent that enhances MLLMs with visual and spatial reasoning tools, demonstrating substantial performance improvements on the benchmark and in robotic experiments. AI

Researchers have introduced a new framework called Think Thrice Before You Speak (TTBYS) to enhance the Theory of Mind (ToM) capabilities in large language models for persuasive dialogue. This framework addresses limitations in current models by explicitly modeling the sequential dependencies among mental states like beliefs and desires, using the Belief-Desire-Intention (BDI) framework. To support this, they also created a large dataset, ToM-based Broad Persuasive Dialogues (ToM-BPD), and demonstrated that a Qwen3-8B model augmented with TTBYS outperformed GPT-5 on predicting mental states and persuasive strategies. AI

Microsoft has integrated GPT-5.5 Thinking and ChatGPT Images 2.0 into its Microsoft 365 Copilot, aiming to enhance its capabilities beyond initial criticisms. This move is part of a broader trend where companies like Meta are also advancing AI-powered hardware, such as their new AI smart glasses developed with EssilorLuxottica, which are seeing increased adoption and upgrades. AI