Unverified70% confidenceFactExact time
Industry estimates suggest GPT-4's training used approximately 25,000 NVIDIA A100 GPUs over several months, with compute costs alone exceeding $100 million
1
Sources
70%
Confidence
Long-term
Relevance
8/3/2026
First Seen
Sources
Related Entities
Related Claims
Verified训练一次GPT-4级别的模型需要数万块A100/H100 GPU协同工作数月85% similarUnverifiedGPT-4于2022年8月训练完成，参数量不到2万亿；当前GPT-5.6约4万亿参数，Fable约10万亿参数，仅为GPT-4的5到6倍79% similarUnverified训练一个GPT-4量级的模型需要数亿美元投入，而调用其API构建应用的边际成本已降至极低76% similarVerifiedGPT-4级别模型训练所需算力相比GPT-3提升了约100倍74% similarUnverifiedGPT-4的参数量据估计超过1万亿，训练过程涉及的浮点运算次数达到10²⁴数量级74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/680113API
curl https://kongchang.com/api/v1/knowledge/claims/680113MCP
get_claim(id=680113)