GPT-4 technical report. Architecture details undisclosed by OpenAI (parameter count, training data, and hardware not published). 8K token context at launch, extended to 128K with GPT-4 Turbo (November 2023). Multimodal: accepts both text and image inputs.

At launch, GPT-4 set new SOTA on virtually every benchmark: scored in the 90th percentile on the bar exam, 86.4% on MMLU (5-shot), and demonstrated strong coding (67% HumanEval). Represented a significant leap over GPT-3.5 in reasoning, factuality, and instruction following. AA Intelligence Index: 13. Proprietary.

Model Details

Architecture DENSE
Context window 128,000

Paper

arXiv: 2303.08774

frontiermultimodalreasoning

Related