What is the long context surcharge for GPT-5.4?

When a session's input exceeds 272,000 tokens, the pricing for the ENTIRE session doubles: input goes from $2.50 to $5.00/M tokens, output increases 50% from $15.00 to $22.50/M tokens. This isn't a progressive surcharge—once you cross the 272K threshold, all tokens in that session are priced at the higher rate.

Is a million-token context window better than RAG?

Depends on use case: Million-token context is better for one-time global analysis requiring full-document perspective (codebase architecture review, complete legal document analysis). RAG is better for frequent queries and ongoing knowledge base services due to cost efficiency—long context triggers 2x pricing while RAG only loads relevant chunks.

How does GPT-5.4 compare to Gemini 2.5 Pro for long context?

Both support 1M+ token contexts. GPT-5.4 is considered slightly stronger on reasoning and code comprehension; Gemini 2.5 Pro is more competitive on pricing for long context use cases and has better multimodal capabilities (video/image understanding). Gemini integrates more deeply with Google Workspace. Choice depends on use case and budget.

— AI DAILY

Sources

OpenRouter
OpenAI Help