According to the first comprehensive benchmarking study published by LegalBenchmarks.ai in September 2025, AI tools in several cases produced contract drafts that were as accurate as, or more accurate than, those prepared by human lawyers. The study evaluated 13 AI tools across 30 real-world contract drafting tasks, where human
xAI announced the Grok-4-Fast model in September 2025, which integrates "reasoning" and "non-reasoning" modes into a unified architecture whilst using 40% fewer thinking tokens than the Grok-4 model. The Grok-4-Fast can process up to 2 million text units at once and has been trained to use
Chinese AI firm DeepSeek released its new 685-billion parameter V3.1 artificial intelligence model on August 19, 2025. This system directly challenges American AI giants, as the model uploaded to the Hugging Face platform achieved a 71.6% score on the Aider coding benchmark. This result surpasses Claude Opus 4&
On August 5, 2025, OpenAI released two open-weight reasoning models under the Apache 2.0 license, named gpt-oss-120b and gpt-oss-20b, allowing researchers to freely access, modify and distribute these AI models. This industry-milestone move responds to growing demand for open-source, high-performance models that make AI development more transparent. The two
On August 7, 2025, Anthropic released Claude Opus 4.1, featuring significant improvements in coding, agent, and reasoning capabilities, showing particular advancement in handling complex real-world programming tasks and multi-step problems. The updated model delivers 38% better performance on coding tasks, and 27% enhanced reasoning capabilities on HumanEval, MMLU, and
AI Outperforms Average Humans in Tests Measuring Emotional Capabilities A recent study led by researchers from the Universities of Geneva and Bern has revealed that six leading Large Language Models (LLMs) – including ChatGPT – significantly outperformed human performance on five standard emotional intelligence tests, achieving an average accuracy of 82% compared
Chinese AI company Moonshot AI unveiled its new Kimi K2 model on July 16, 2023, which outperforms OpenAI's GPT-4 in several key performance metrics while being completely free to use. The company, which has previously developed acclaimed AI models, has achieved a significant breakthrough for Chinese AI technology&
LEXam, published on the Social Science Research Network (SSRN) platform, is the first comprehensive benchmark specifically measuring legal reasoning abilities of AI models using 340 authentic legal exam questions. Developed by researchers, the testing system covers regulatory frameworks from six different jurisdictions (United States, United Kingdom, France, Germany, India, and
Officially released on June 26, 2025, Gemma 3n includes significant developments specifically targeting on-device AI operation. The multimodal model natively supports image, audio, video, and text inputs and is available in two sizes: E2B (5 billion parameters) and E4B (8 billion parameters), operating with just 2GB and 3GB of memory
On July 2, 2025, researchers from Helmholtz Munich published the groundbreaking Centaur model in Nature, capable of predicting and mimicking human behaviour across various situations. Led by Marcel Binz, the team created the Psych-101 dataset containing over 10 million decisions from 60,000 participants across 160 psychological experiments, then used
Google unveiled an updated preview version of Gemini 2.5 Pro on 5th June, which achieved a 24-point Elo rating increase on the LMArena leaderboard, maintaining its leading position with 1470 points. On the WebDevArena leaderboard, it rose to 1443 points with a 35-point Elo increase, securing the top position.
ChatGPT, OpenAI's popular AI chatbot, suffered a decisive defeat against a 46-year-old Atari 2600 chess program when researcher Stephen Cobb pitted the two systems against each other in March 2024. During the experiment, the gaming console won five out of six matches, with ChatGPT securing only a single
