Tag archive
How to build browser-use agents on remote browsers that remain reliable as frontier models such as GPT-6 Sol change speed, reasoning, and tool behavior.
OpenAI cut GPT-6 Sol and Luna prices in half, improved caching for agent workloads, and published alignment numbers worth reading before granting an agent more autonomy. Here is what changed, what to verify yourself, and where OpenAI's own benchmark comparisons need a caveat.
Frontier launches are usually capability stories. Tuesday's were pricing stories. Anthropic opened with Claude Opus 5.5, positioned as Fable 5.1-class...
Les lancements frontières racontent d'ordinaire une histoire de capacités. Ceux de mardi étaient une histoire de prix : Claude Opus 5.5 contre GPT-6 Sol et
Week of Sep 18-24: OpenAI expands GPT-6 with Sol and Luna, Anthropic launches Opus 5.5, Grok 4.7 targets agentic coding, and Copilot adds local sandboxing and tracing.