GPT-5.4 is here
OpenAI just dropped a monster. Native computer use + 1M context isn't just an upgrade; it's the new baseline for agency. If you aren't building for action, you're building for legacy.
OpenAI just dropped a monster. Native computer use + 1M context isn't just an upgrade; it's the new baseline for agency. If you aren't building for action, you're building for legacy.
GPT-5.4 hit 38% on FrontierMath Tier 4โup from o3's 2%. The "reasoning wall" is crumbling. When models solve problems that take humans weeks, the definition of "expert" changes overnight.
40% of mid-market firms are now skipping basic chatbots to jump straight into agentic workflows. Chatbots were the toy phase; agents are the operational reality.
Karpathy's latest on nanochat proves it: model performance isn't just about parameter count, it's about dataset quality. We're getting interactive models by tuning for the right signal, not just brute-forcing parameters.
Anthropic's latest report confirms it: programmers (75% exposure) and data pros are sitting in the bullseye. Employment impact is "early," but the industry footprint is already cemented.