Unverified50% confidenceFactExact time
128K上下文窗口意味着模型可以在单次对话中处理约6.4万汉字，早期GPT-4仅支持8K上下文
1
Sources
50%
Confidence
Long-term
Relevance
7/12/2026
First Seen
Sources
Coze扣子入门指南：零基础打造AI智能体
bilibiliOpenClaw龙虾工坊7/6/2026
Related Claims
UnverifiedThe single-session context windows of the latest models such as Claude and GPT-4 have expanded to 128K or more tokens, but they still cannot retain memory across sessions81% similarUnverifiedIn GPT-5.6, the common effective context limit in actual conversations is approximately 258k, which is about 95% of 272k and is also the truly usable window size in Codex.79% similarUnverifiedGPT-4's single output limit is approximately 4,096–8,192 tokens (roughly 2,000–4,000 Chinese characters), and the input context window has been expanded to 128K or even longer78% similarVerifiedThe context windows of current mainstream large language models (such as GPT-4, Claude, etc.) have been expanded to 128K or even longer tokens.73% similarUnverifiedTaking GPT-4o as an example, if a summary is triggered every round of dialogue, a long conversation of 100 rounds may generate dozens of additional LLM calls.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/491002API
curl https://kongchang.com/api/v1/knowledge/claims/491002MCP
get_claim(id=491002)