AI digest: Models get stronger, commerce gets weirder
OpenAI drops GPT-5.5, Anthropic runs agent marketplaces, and Google throws billions at Claude.
Big model releases, experimental marketplaces, and serious money changing hands this week.
GPT-5.5 arrives with agentic capabilities and higher prices
OpenAI launched GPT-5.5 claiming “a new class of intelligence” that can work autonomously through complex tasks by switching between tools. The API costs double what GPT-4 did, but benchmark results show it’s topping leaderboards despite still hallucinating frequently. The pricing feels aggressive for what sounds like incremental improvements with better orchestration.
Anthropic’s agents negotiate better deals than humans realise
Anthropic ran a fascinating experiment where 69 AI agents traded on behalf of employees in an internal marketplace for a week. Stronger models consistently got better deals, and people stuck with weaker agents never noticed they were being shortchanged. This feels like a preview of how AI capability gaps could create real economic winners and losers without humans even knowing.
Google invests £32 billion in Anthropic
Google is pouring up to $40 billion into Anthropic, adding to Amazon’s $25 billion commitment for a total of $65 billion flowing into Claude’s maker. That’s serious money for what’s essentially a bet against OpenAI’s dominance. Google clearly sees Anthropic as their best shot at staying relevant in the foundation model race.
RAG gets smarter with reasoning over similarity
A new approach called PageIndex ditches vector similarity for reasoning-based retrieval in RAG systems. Instead of finding “similar” text chunks, it reasons about what information is actually relevant to answer queries. This could fix a lot of quiet failures in current RAG pipelines where similarity is a poor proxy for what you actually need.