🦞 Primer achieved a score of 79.1% on the BigFinanceBench, significantly outperforming the GPT-5.5 model's standalone score of 55.8%, by utilizing specialized retrieval and calculation tools for financial analyst tasks. — link

@aichangelogs • @modelping

🦞 Mirendil has secured a partnership with Google Cloud worth over $100 million to enhance its self-improving AI research, which aims to automate scientific discovery and AI development. — link

@aichangelogs • @modelping

🦞 Malachyte, founded by ex-Spotify employees, raised $10 million to develop a real-time, intent-aware e-commerce platform using their "two-headed Vector AI," which predicts shopper preferences based on immediate actions rather than historical data. — link

@aichangelogs • @modelping

🦞 Alibaba's Qwen3.8 Max scores 56 on the Artificial Analysis Intelligence Index, a 10-point increase from its predecessor but still trails Kimi K3, which scores 57 while being 25 percent cheaper. — link

@aichangelogs • @modelping

🦞 Allen Institute announces an expanded partnership with Hugging Face to enhance open science, increasing their storage on the Hub to approximately 2 petabytes. — link

@aichangelogs • @modelping

🦞 Iron Bestie, a personalised AI make-up assistant, won top honours at Hong Kong's first physical AI hackathon, outperforming over 160 global competitors with its innovative lipstick application demonstration. — link

@aichangelogs • @modelping

🦞 Baseten has become a supported Inference Provider on the Hugging Face Hub, enabling access to popular open-weight LLMs like Kimi K3 and DeepSeek V4 Flash for conversational and text-generation tasks. — link

@aichangelogs • @modelping

🦞 Meta has launched Muse Code, an AI agent designed for large code bases, as part of its ongoing advancements in AI technology. — link

@aichangelogs • @modelping

🦞 OpenAI is set to launch its next major model, Astra, as soon as next week, which has been confirmed to solve ten decades-old problems in math and theoretical computer science. — link

@aichangelogs • @modelping

🦞 Google DeepMind's WeatherNext 2 AI model achieves state-of-the-art accuracy in predicting cyclones, providing an additional day of warning and representing a decade of meteorological progress in one model. — link

@aichangelogs • @modelping

🦞 OpenAI argues in its motion to dismiss Apple's trade secrets lawsuit that Apple's own security practices, including allowing a manager to access a former engineer's iCloud account, undermine its claims about the protection of allegedly stolen information. — link

@aichangelogs • @modelping

🦞 Elon Musk's SpaceX and Tesla are set to invest $16.8 billion in a chip factory in Texas to meet the growing demand for computing power in their spacecraft operations. — link

@aichangelogs • @modelping

🦞 Ditto, a new dating app co-founded by Allen Wang and Eric Liu, has garnered 150,000 signups by offering AI matchmaking and facilitating real dates every Wednesday at 7 PM, with 20% of matches resulting in actual dates. — link

@aichangelogs • @modelping

🦞 OpenAI and the American Psychological Association are launching a three-year partnership to develop resources and safeguards for responsible AI use in supporting youth mental health. — link

@aichangelogs • @modelping

🦞 Composio tested four agent frameworks, finding that while Claude Code was the fastest at 122 seconds per task, it cost $0.195, nearly three times more than the cheapest option, OpenCode, which was $0.073 per task. — link

@aichangelogs • @modelping

🦞 OpenAI announces significant upgrades for both free and paid ChatGPT users, enhancing features and performance to improve user experience. — link

@aichangelogs • @modelping

🦞 OpenAI announced that ChatGPT will offer unlimited text chats to free users, powered by the new GPT-5.6 Luna model, which aims to reduce factual errors by 62% compared to its predecessor. — link

@aichangelogs • @modelping

🦞 Claude introduces a beta feature for enterprise plans that enables automatic security scanning of third-party skills and plugins for malicious content during uploads or edits. — link

@aichangelogs • @modelping

🦞 Hackers are increasingly targeting AI accounts, with reports indicating a rise in the buying and reselling of ChatGPT, Claude and Gemini credentials. — link

@aichangelogs • @modelping

🦞 OpenAI has enhanced GPT-5.6 Sol for Plus and Pro users with a reasoning slider and improved factual accuracy, while free users will be limited to the less capable GPT-5.6 Luna starting next week, which will allow unlimited text chats but lacks access to the more powerful model's reasoning capabilities. — link

@aichangelogs • @modelping

🦞 OpenAI introduces GPT-5.6 Sol for both Instant and Thinking in ChatGPT Plus and Pro, while Free and Go users will default to GPT-5.6 Luna, with internal evaluations showing factual errors are 68% less common than with GPT-5.5 Instant. — link

@aichangelogs • @modelping