AI News Roundup: Beam 501B Open Model, GPT-Rosalind, Stack Overflow for Agents & More — 16 October 2026

Here is everything worth knowing in AI today, 16 October 2026.

Beam: A 501-Billion-Parameter Open Model You Can Own and Customise

Hugging Face has announced Beam, a 501-billion-parameter open model that organisations can own, customise, and scale on their own infrastructure. The release is positioned as a major step for the open-model ecosystem, giving enterprises and researchers genuine ownership of a frontier-scale model rather than relying solely on API access. Details on architecture and training data are expected to follow in accompanying technical documentation, but the headline figure — 501B parameters — places Beam firmly in the territory of the largest openly available models to date.

OpenAI Launches GPT-Rosalind Academic Plan for Life Sciences Labs

OpenAI has introduced the Rosalind Academic Plan, offering eligible academic research laboratories sponsored access to GPT-Rosalind and a suite of tools aimed at advancing life sciences research. The plan is named after the pioneering scientist Rosalind Franklin and is designed to lower the barrier for university and research-institute labs to apply large language models to biological and chemical problems. This follows a broader industry trend of AI companies creating discounted or sponsored tiers specifically for academic users.

Stack Overflow for Agents Sees Up to 10× Growth via ChatGPT Plugin

OpenAI has published a case study showing that two engineers built Stack Overflow for Agents in just a few weeks using Codex, and that integrating a ChatGPT plugin drove up to ten times growth in agent creation, with sign-ups tripling after the plugin launched. The story illustrates how AI-native developer tools, combined with distribution through ChatGPT's plugin ecosystem, can accelerate both product development and user adoption at a pace that would have been difficult to achieve with conventional software development cycles.

Postgres 19 Release Delayed After AI Reviewers Surface More Bugs Than Developers Can Fix

In a striking example of AI creating unexpected bottlenecks, The Register reports that the release of Postgres 19 has been delayed because AI-powered code reviewers are identifying bugs faster than the human developer community can address them. Rather than speeding up the release cycle, the AI tooling has produced a backlog of confirmed issues that the project's maintainers must work through before shipping. The story, which gained significant traction on Hacker News, raises practical questions about how open-source projects manage quality pipelines when AI reviewers operate at a speed that outpaces human capacity to act on their findings.

Research: 90% of US Working Hours Are on Tasks AI Benchmarks Don't Cover Well

New research from Mercor, highlighted on Hacker News with considerable engagement, finds that roughly 90 per cent of US working hours are spent on tasks that current AI benchmarks do not measure well. The finding is a pointed reminder that benchmark performance — the standard metric used to compare AI models — may be a poor proxy for real-world economic impact. The research suggests that the gap between what AI can demonstrably do in controlled evaluations and what it can reliably replace in day-to-day work remains substantial.

IBM Releases Granite Switch 4.1 — Chat, Decisions, and Speech from One Endpoint

IBM has released Granite Switch 4.1 via Hugging Face, a unified model that handles chat, decision-making, and speech tasks through a single API endpoint. Consolidating multiple modalities and task types behind one endpoint simplifies deployment for enterprise teams that would otherwise need to manage separate models for different use cases. The release continues IBM's strategy of building production-ready open models under the Granite family for regulated and enterprise environments.

Sources and links for all items are listed below.