GPT 5: Much Ado About Nothing

There’s nothing about GPT 5 that makes me want to stop using Claude Opus 4.1 or even Gemini 2.5 Pro (I don’t use Grok 4). LLM performance has definitely hit a wall (I don’t use Claude Code, I prefer to piece the code together myself), so now what?

Those $100 billion models are not happening in 2026, or at least not at the valuations that OpenAI and Anthropic had hoped. If the U.S. Government subsidizes the buildout I will be pissed. It would simply be another scam like David Sacks and the Trump Admin’s money grab with Crypto issuance and regs.

LLM innovation is going to have to come from the inference side, because you simply can’t keep “10-xing” the amount of spend on the training side, which implies a Trillion dollar LLM cost in 2027/28 after the $100 billion LLM build cost in 2026.

Net, net, LLMs - Anthropic’s Claude family in particular - provide a significant opportunity for Software companies to capture material Operating Margin lift on the R&D line - Enterprise Software companies in particular.

See our related articles here and here.

---