GPT-4.5 (discontinued June 2026) is kept on this board as a writing reference, not as a general assistant. Our tests still put its creative writing at the top—voice, rhythm, and tone-sensitive drafting that later GPT-5.x models have not matched. Everything else is weaker: slower and far more expensive than modern models, with middling helpfulness, instruction following, and comprehension for everyday tasks. Use the score as a reminder of what got traded away when labs optimized for agents—not as a recommendation to run it for work.