Unverified50% confidenceFactExact time
200,000 lines of code far exceeds the window limits of tens of thousands of tokens in earlier models.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Gemini 4 Argon Optimizes 200,000 Lines of Code Overnight, Tops Three Benchmarks
bilibiliAI每日速读10/14/2026
Related Entities
Related Claims
UnverifiedEarly mainstream models had context windows of only 4,000 to 8,000 tokens, while models from the past two generations have expanded the window to 100,000 tokens or higher.75% similarUnverified早期Codex模型（2021年）的上下文窗口仅有4K Token，约相当于200行代码75% similarUnverifiedThe context window refers to the upper limit of tokens a model can process simultaneously in a single inference; 128K is roughly equivalent to a medium-length book, and 1 million tokens is roughly equivalent to thousands of pages of documents or a small-to-medium-sized codebase.74% similarVerifiedA context window of 200K tokens can ingest approximately 150,000 lines of code at once, equivalent to the entire source code of a medium-sized commercial project.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1041707API
curl https://kongchang.com/api/v1/knowledge/claims/1041707MCP
get_claim(id=1041707)