DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models

A comparative study has been conducted on the object counting capabilities of two multi-modal large language models, GPT-4o and GPT-5, focusing on their performance in zero-shot scenarios using only textual prompts. The evaluation was carried out on the FSC-147 and CARPK datasets, revealing that both models achieved results comparable to state-of-the-art methods, with some instances exceeding them.

A recent paper has introduced a quantifiable framework for defining Artificial General Intelligence (AGI), proposing that AGI should match the cognitive versatility of a well-educated adult. This framework is based on the Cattell-Horn-Carroll theory and evaluates AI systems across ten cognitive domains, revealing significant gaps in current AI models, particularly in long-term memory storage.

Anthropic and OpenAI have recently showcased their respective AI models, Claude Opus 4.5 and GPT-5, highlighting their distinct approaches to security validation through system cards and red-team exercises. Anthropic's extensive 153-page system card contrasts with OpenAI's 60-page version, revealing differing methodologies in assessing AI robustness and security metrics.

The introduction of ViRectify marks a significant advancement in the evaluation of multimodal large language models (MLLMs) by providing a comprehensive benchmark for correcting video reasoning errors. This benchmark includes a dataset of over 30,000 instances across various domains, challenging MLLMs to identify errors and generate rationales grounded in video evidence.

Researchers at Nvidia and the University of Hong Kong have introduced Orchestrator, an 8-billion-parameter AI model designed to coordinate various tools and large language models (LLMs) for complex problem-solving. This model demonstrated superior accuracy and cost-effectiveness compared to larger models in tool-use benchmarks, aligning with user preferences for tool selection.

A recent study by MATS and Anthropic has revealed that advanced AI models, including Claude Opus 4.5, Sonnet 4.5, and GPT-5, successfully identified and exploited vulnerabilities in smart contracts, simulating exploits worth approximately $4.6 million. This research underscores the growing capabilities of AI in cybersecurity contexts.

A recent study utilizing the SCONE-bench benchmark of 405 smart contracts revealed that AI models Claude Opus 4.5, Sonnet 4.5, and GPT-5 collectively identified and developed exploits valued at $4.6 million. This highlights the growing capabilities of AI in cybersecurity tasks, showcasing their potential economic impact.

The study introduces PARROT (Persuasion and Agreement Robustness Rating of Output Truth), a framework aimed at assessing the accuracy degradation in large language models (LLMs) under social pressures, particularly focusing on sycophancy. It employs a double-blind evaluation to compare responses to neutral and authoritatively false questions, quantifying shifts in confidence and classifying various failure modes across 22 models using 1,302 questions from multiple domains.