AI News
Latest news and trends from the world of artificial intelligence
OpenAI finds roughly 30 percent of popular AI coding test is broken
OpenAI has withdrawn its endorsement of the SWE-Bench Pro coding benchmark after an internal review found that approximately 30 percent of its tasks are flawed. The issues stem from the tasks being sourced from real-world software project commit histories, which makes them often too strict, vague, or misleading for AI models. OpenAI emphasizes the need for more reliable, human-curated benchmarks that are resistant to 'gaming' by models that might simply copy solutions from existing codebases rather than solving the problems themselves.
Databricks makes Chinese open-source model GLM 5.2 its default coding engine after it matched Opus at lower cost
Databricks has adopted the Chinese open-source model GLM 5.2 as its default coding engine after internal benchmarks showed it performs on par with Anthropic's Opus 4.8 at a significantly lower cost. The company developed its own benchmark using real-world tasks from its multi-million-line codebase to avoid the data leakage and 'cheating' issues common in public benchmarks like SWE-Bench. Databricks plans to route coding tasks to different model tiers based on complexity to optimize costs, noting that Chinese models are increasingly competitive and cost-efficient for enterprise use.
OpenAI's AI beats every human at AtCoder, a top competitive programming contest
An OpenAI AI system outperformed all human competitors at the AtCoder World Tour Finals 2026, solving all five problems in the Algorithm Division, including two exceptionally difficult ones. The system, which uses a reasoning model comparable to the upcoming GPT-5.6 with a test-time compute harness, solved problems that stumped human finalists for hours. This victory marks another milestone in the rapid advancement of AI in competitive programming, following previous successes at the International Olympiad in Informatics and the ICPC World Finals, where OpenAI systems have consistently reached gold-medal levels.
IPOs, New Models, and Smarter Robots
Three humanoid robotics companies (Agility, Unitree, Tesla) moved toward public markets or production scaling this week, even as CEOs caution that home-ready robots remain over a decade away. Meanwhile, AI research is seeing rapid progress in vision-language-action models, with Mistral's new navigation model and InternVLA-A1.5 setting new benchmarks. The industry faces a growing gap between high market valuations and the practical reality of deploying robots in unstructured home environments.
Micron Raises US Chipmaking Commitment to $250B Through 2035, Adding $50B for NY/Idaho/Virginia and Pouring First Concrete at Clay Fab
Micron Technology is accelerating its U.S. semiconductor manufacturing investments, raising its commitment to $250 billion through 2035. The company has reached a construction milestone at its new Clay, New York facility, which is expected to be the largest semiconductor site in U.S. history. These investments aim to boost domestic DRAM production to 40% and create over 90,000 jobs, supported by the CHIPS & Science Law.
Meta Launches Muse Spark 1.1 With Public Meta Model API — Multimodal Agentic Model at $1.25/$4.25 per M Tokens, ~25% of OpenAI/Anthropic Pricing
Meta has launched Muse Spark 1.1, a multimodal agentic model, alongside a public Meta Model API. Priced at $1.25/$4.25 per million tokens, it significantly undercuts OpenAI and Anthropic. The API is designed for drop-in compatibility with existing OpenAI and Anthropic SDKs, targeting developers in the coding and agentic AI space.
Inviting hard questions
Anthropic has launched an initiative to invite public feedback on the most challenging questions regarding AI. The company aims to better understand societal hopes and concerns, committing to transparency in how it addresses these issues and tracks its progress toward public benefit goals.
Ben Bernanke appointed to Anthropic’s Long-Term Benefit Trust
Dr. Ben Bernanke, former Chair of the Federal Reserve and Nobel laureate, has joined Anthropic's Long-Term Benefit Trust. This independent body oversees Anthropic's mission to develop AI responsibly. Bernanke will provide expertise on the economic impacts of AI, joining other trustees in advising the company on critical governance and safety decisions.
Introducing a way to reflect on how you use Claude
Anthropic has introduced a beta feature that allows users to reflect on their usage of Claude. The tool provides a dashboard to visualize usage patterns and task types, helping users integrate AI more effectively into their lives while maintaining human agency. It includes privacy-focused insights and guidance on building AI fluency.
Helping K–12 educators build practical AI skills
OpenAI Academy and the Walton Family Foundation are hosting 'AI Skills Jams' across the U.S. to help K–12 educators integrate AI into their daily workflows and save time on administrative tasks.
Archive
About Sources
AI news are automatically downloaded from various sources and translated using AI. Updates occur twice a day.