GPT-4.5 (Orion)
OpenAI's largest-ever trained model, scaled on unsupervised learning before reasoning took over
Retired
GPT-4.5 (Orion) appears to be discontinued or shut down — pulled from API Jul 2025; full sunset mid-2026 This page is kept as an archive record and is excluded from search engines.
About
GPT-4.5, codenamed Orion, was OpenAI's largest model by training compute when it launched on February 27, 2025. It doubled down on scaling unsupervised pretraining rather than test-time reasoning, and it showed: stronger world knowledge, better writing, and noticeably more emotionally intelligent responses than GPT-4o. But it was not a frontier reasoning model, and at $75/$150 per million tokens it was brutally expensive to serve. OpenAI pulled it from the API on July 14, 2025 in favor of the cheaper GPT-4.1, then removed it from ChatGPT when GPT-5 launched, with a full sunset announced for mid-2026. It remains an interesting historical marker: the last big bet on pure pretraining scale before the industry pivoted to reasoning.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Team-reported: outperformed GPT-4o on MMLU across 15 languages
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Feb 27, 2025 release, Orion codename, $75/$150 pricing verified via Wikipedia + TechCrunch; API removal July 14, 2025 and ChatGPT sunset timeline documented; Largest-ever-trained claim attributed to OpenAI; not-frontier-level caveat captured; Honest framing: historical marker for end of pure-pretraining scaling era.