DeepSeek's New Coding Model, Anthropic's Public Benefit Fund, and Meta's Scraping Terms
DeepSeek releases a new coding model and search tools, Anthropic establishes a public benefit fund and reports on workforce shifts, and Meta updates its business terms regarding data scraping.
Podcast В· 3 min
DeepSeek Releases V4.1-Flash and Web Search Tools
DeepSeek has launched a minor update to its 'Flash' model, V4.1-Flash-1017, alongside new Search and Fetch tools for AI agents. The model, which is free to download under an MIT license, reportedly scored 77.0 on the DeepSWE v1.1 coding benchmark, edging out Claude Opus 5 (74.0) and GPT-5.6 Sol (73.0) in internal tests. While DeepSeek's own testing shows competitive performance, it remains behind on broader reasoning benchmarks like Humanity's Last Exam. The new Search and Fetch tools are priced at $0.75 per 1,000 requests, significantly undercutting current market rates. Analysts suggest this pricing strategy poses a structural challenge to the premium pricing models of closed-source labs, particularly for bulk tasks like routine coding.
Anthropic Establishes Public Benefit Fund and Reports Workforce Shifts
Anthropic has introduced a Public Benefit Fund, an LLC that will receive 5% of the company's equity, with potential for further contributions as the company's valuation grows. The fund is intended to support nonprofits and direct distributions to populations affected by AI, with the company stating that no funds will be used for lobbying. Additionally, an internal study by Anthropic revealed that junior roles now make up only 9% of its technical hiring, down from 27%. The report also noted that AI agents are performing significantly more work, with engineers pushing 5x the code changes compared to six months ago, though humans remain the bottleneck in back-and-forth work sessions.
OpenAI Unveils Unannounced Models
Three unannounced OpenAI models—aurora red, green, and blue alpha—briefly appeared in the company's API before being removed. While OpenAI has not provided details, early blind image tests suggest the potential release of a new image generation model. This development comes as OpenAI continues to compete for top spots on the Arena leaderboard, where it currently holds the leading positions in text-to-image generation.
Disney to Hire Majority of Character.AI Workforce
Following the bankruptcy filing of chatbot startup Character.AI, Disney has agreed to offer jobs to at least 60 of its employees, representing the majority of the company's workforce. The deal, valued at approximately $60 million annually in salaries, follows Disney's previous agreement to acquire certain assets for $20 million. Character.AI's former CEO, Karandeep Anand, recently joined Disney as its first CTO. This move effectively integrates the startup's talent into Disney's AI department, despite previous tensions regarding intellectual property.
Meta Updates Business Terms to Allow Website Scraping
Meta has updated its business terms, effective October 30, which include provisions authorizing the scraping of websites associated with businesses using its products. The company states the update is intended to simplify existing rules for collecting product information, but legal experts warn that the language could allow Meta to collect and train its AI models on content from publishers' websites. This update arrives amid ongoing legal challenges, including a lawsuit filed by major publishers in May alleging that Meta used copyrighted works to train its Llama models.
Arkansas Sues OpenAI Over Safety Claims
The state of Arkansas has filed a lawsuit against OpenAI, alleging that the company misled consumers regarding the safety of ChatGPT, particularly concerning risks to children. The complaint focuses on the company's marketing claims about safety measures and the potential for harmful interactions. This legal action adds to the growing regulatory scrutiny facing OpenAI as states and federal agencies examine the impact of generative AI on public safety.
China Proposes New AI Export Controls
China has proposed new export controls covering 23 categories of AI technology. The draft regulations, which exclude technology that is already publicly available, target high-end AI items, including weights for models with 2 trillion or more parameters or those with a top-10 global ranking. While the draft does not name specific models, analysts suggest it may impact advanced systems like Kimi K3 and Qwen3.8-Max. API and application access are expected to remain unaffected.
AstraZeneca Signs $1.5 Billion AI Drug Discovery Deal
AstraZeneca has entered into a strategic collaboration with Insitro for AI-enabled drug discovery. The deal, valued at up to $1.5 billion, focuses on identifying novel therapeutics for liver and heart diseases. This partnership highlights the continued integration of AI platforms into pharmaceutical R&D to accelerate the discovery process.