Krux
March 16, 2026
OpenAI's GPT-5.4 Reads 1 Million Tokens at Once
Published: March 16, 2026 at 12:55 AM
Updated: March 16, 2026 at 12:55 AM
100-word summary
OpenAI released GPT-5.4 on March 5 with a context window stretching to 1 million tokens. That's enough to ingest roughly 750,000 words in one go, turning the model into something closer to a research assistant that's already read every relevant document before you ask your first question. Two new modes ship alongside: Thinking, which exposes chain-of-thought reasoning and claims to reduce deceptive outputs, and Pro, tuned for higher-performance tasks. A new Tool Search feature lets the API automatically find and use the right tools across complex workflows. The model also cuts token costs compared to earlier versions, making it cheaper to run those million-token marathons.
What happened
OpenAI released GPT-5.4 on March 5 with a context window stretching to 1 million tokens. That's enough to ingest roughly 750,000 words in one go, turning the model into something closer to a research assistant that's already read every relevant document before you ask your first question. Two new modes ship alongside: Thinking, which exposes chain-of-thought reasoning and claims to reduce deceptive outputs, and Pro, tuned for higher-performance tasks. A new Tool Search feature lets the API automatically find and use the right tools across complex workflows.
Why it matters
The model also cuts token costs compared to earlier versions, making it cheaper to run those million-token marathons.