Today's AI news: Claude Haiku 5.5 beats GPT-6 Luna, OpenAI solves 90 top maths problems, revenue shortfall, Meta & Microsoft limit Claude use.
Claude Haiku 5.5 Outperforms GPT-6 Luna at the Same Price
Anthropic's latest small model, Claude Haiku 5.5, has landed — and according to AINews at Latent Space, it performs better than OpenAI's GPT-6 Luna while sitting at the same price point. The release is a notable win for the efficient-model camp, signalling that smaller, cheaper models are continuing to close the gap on their larger rivals. The news comes as competitive pressure across the model tier intensifies, with every major lab racing to deliver more capability per dollar.
OpenAI Publishes 722 Maths Papers, Solving 90 of the World's Top 500 Open Problems
In what is being described as possibly the most significant moment in over a century of mathematics, OpenAI has published 722 papers tackling open mathematical problems — solving 90 of the top 500 outstanding problems in the field. The release, dubbed the Quasi-Riemann-Hypothesis moment by the AINews team, represents an extraordinary demonstration of AI-driven scientific discovery. The scope and ambition of the output have left the mathematics community reeling, and the publication has quickly become one of the most talked-about AI research events in recent memory.
OpenAI's Annualised Revenue $20 Billion Below Earlier Signals
A significant financial story broke today via CNBC and surfaced prominently on Hacker News: OpenAI's annualised revenues are running approximately $20 billion below figures that had previously been signalled to the market. The report, which also mentions Nvidia, Oracle, and CoreWeave in context, is likely to raise questions about the pace of the company's commercial growth relative to its spending on compute and talent. The story was one of the most-discussed items on Hacker News today, accumulating 225 points.
Meta and Microsoft Move to Limit Employee Use of Claude
Both Meta and Microsoft have reportedly taken steps to reduce their employees' use of Anthropic's Claude AI, according to a widely circulated report that became the most-upvoted AI story on Hacker News today (363 points). The move is notable given that both companies have their own large AI investments — Meta with its Llama model family and Microsoft through its deep partnership with OpenAI. The reasons behind the restrictions have not been detailed in the available reporting, but the development underscores the increasingly competitive and politically sensitive nature of AI tool adoption inside major tech firms.
OpenAI Disrupts AI-Enabled Influence Operations
OpenAI announced it has disrupted two AI-enabled influence operations that used false-front journalists and a fabricated think tank to spread geopolitical messaging. The operations leveraged AI to manufacture a veneer of credibility — creating fake personas and institutional fronts to push specific narratives. OpenAI's action is a reminder that AI safety concerns extend well beyond model outputs to include deliberate misuse by state or non-state actors seeking to manipulate public discourse.
StepFun's Step 5 Preview — a 1M-Context MoE — Appears on OpenRouter
A model called Step 5 Preview, developed by StepFun and described as a mixture-of-experts (MoE) architecture with a one-million-token context window, has quietly appeared on the OpenRouter platform. The listing attracted attention on Hacker News (53 points), with researchers and developers noting the enormous context length as a potentially significant capability for long-document and multi-turn applications. Little else has been officially announced by StepFun at this stage, but the model's appearance on a public routing platform suggests it is available for testing now.
Frequently Asked Questions
How does Claude Haiku 5.5 compare to GPT-6 Luna?
According to AINews at Latent Space, Claude Haiku 5.5 performs better than OpenAI's GPT-6 Luna while being offered at the same price point, making it a strong option in the small, efficient-model category.
What did OpenAI achieve in mathematics?
OpenAI published 722 mathematics papers and solved 90 of the world's top 500 open mathematical problems. The AINews team described it as potentially the most significant moment in over a century of mathematics.
Why are Meta and Microsoft limiting employee use of Claude?
Both Meta and Microsoft have reportedly taken steps to reduce their employees' use of Anthropic's Claude AI. The specific reasons have not been detailed, but the move highlights competitive tensions in enterprise AI tool adoption.
What is OpenAI's revenue situation as of October 2026?
According to a CNBC report, OpenAI's annualised revenues are running approximately $20 billion below figures that had previously been signalled to the market, raising questions about its commercial growth trajectory.