PlainAI AI explained plainly

OpenAI · 2025 · 02 · 27 · Model · ~1 min read

OpenAI launched GPT-4.5 (Orion)

OpenAI's biggest, most expensive non-thinking model — codename 'Orion'. Smoother writing, fewer made-up facts, and very pricey. OpenAI itself called it underwhelming relative to cost, then quietly retired it within months.

What's actually new

  • Largest pre-training run OpenAI had done at the time it launched.
  • Better at vibes — emotional intelligence, creative writing, fewer 'I cannot help with that' refusals.
  • Lower made-up-fact rate on factual questions vs GPT-4o.

If you want more

Worth knowing~30s
  • Eye-watering price. $75 per million input tokens, $150 per million output tokens at launch — about 30x GPT-4o.
  • Retired within months. OpenAI deprecated GPT-4.5 once GPT-5 shipped, citing 'not justifying its cost'.
  • The Orion saga ended the 'just bigger pre-training' era. Everyone moved to thinking-style scaling instead.
Who should care~20s

Anyone curious about why OpenAI shifted away from pre-training scaling. Researchers studying scaling laws. Developers who briefly tried GPT-4.5 and got the bill.

What to do about it~20s

Read the post-mortem analysis from people like Dario Amodei and Sam Altman about why bigger pre-training stopped paying off. The story matters more than the model itself.

Honest take~45s

GPT-4.5 was the model that proved 'just train bigger' had hit its ceiling. The market had been betting on this paradigm to keep working forever; it didn't. OpenAI's pivot to thinking-style scaling (the o-series) became the default for everyone. Orion's quiet retirement was the official funeral for the 'bigger is always better' assumption — and the moment AI labs all started solving capability through other tricks.

Other recent model updates

Sources

Last verified · 2026 · 05 · 05 · Found a fact wrong? corrections@aguidetocloud.com