Unverified50% confidenceFactExact time
GPT-4的上下文窗口从8K扩展到128K乃至百万级别，上下文窗口越大每次调用消耗的令牌越多
1
Sources
50%
Confidence
Long-term
Relevance
7/18/2026
First Seen
Sources
Related Claims
Unverified早期Codex的上下文窗口约为4096个token，对应大约150-300行代码；GPT-4系列将这一限制扩展至128K token70% similarUnverified128K上下文窗口的注意力矩阵规模是32K窗口的16倍70% similarUnverifiedHunyuan V3 has a context window of 256K and a maximum output of approximately 8K.70% similarVerifiedThe context windows of current mainstream models are approximately 128K to 1 million tokens.69% similarUnverifiedDeepSeek V4's maximum output can reach 384K tokens, while GPT-5.5's is 128K tokens.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/553506API
curl https://kongchang.com/api/v1/knowledge/claims/553506MCP
get_claim(id=553506)