Unverified50% confidenceFactExact time
128K 字符的输出上限约相当于 3-4 万个 token
1
Sources
50%
Confidence
Long-term
Relevance
9/6/2026
First Seen
Sources
Related Entities
Related Claims
Unverified15,000行代码换算成Token大约在150,000到300,000之间，取决于代码密度与注释比例69% similarUnverified在单张 32GB 显卡、q4_0 量化场景下，上下文长度从更新前的 82,432 token 增至更新后的 199,680 token68% similarUnverified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 token68% similarUnverified最新模型的上下文长度已达128K token67% similarUnverified170k tokens场景下，注意力得分矩阵在BF16精度下约需54GB显存（170000×170000×2字节）67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/864896API
curl https://kongchang.com/api/v1/knowledge/claims/864896MCP
get_claim(id=864896)