What happened
OpenAI released GPT-5.4 in three variants — Standard, Thinking, and Pro — supporting context windows up to 1.05 million tokens. The model reduces individual claim errors by 33% and full-response errors by 18% compared to GPT-5.2. Concurrently, OpenAI retired GPT-5.1 models from ChatGPT, auto-migrating conversations to 5.3 or 5.4 variants.
Why it matters
GPT-5.4 is an iterative improvement rather than a paradigm shift. The three-variant approach (Standard/Thinking/Pro) mirrors what other labs already offer, and the accuracy improvements, while meaningful, are incremental. The 1.05M context window brings OpenAI to parity with Anthropic and Google rather than establishing a lead. The GPT-5.1 deprecation signals OpenAI's increasingly aggressive model rotation cycle.
Who should pay attention
Teams currently on GPT-5.1 or 5.2 who need to plan migration, developers comparing frontier model pricing and capabilities across labs, and ChatGPT API users who should test their prompts against 5.4 for regressions.