Unverified50% confidenceBenchmarkExact time
The same code refactoring task may consume 3,000 tokens in standard mode and may balloon to over 50,000 tokens in Extra High mode.
1
Sources
50%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
Codex超高推理模式隐藏成本：一个任务耗尽70%额度的真相
redditr/OpenAI7/12/2026
Related Claims
UnverifiedLarge language models have an inherent context window limitation, with the number of tokens that can be processed in a single pass typically ranging from tens of thousands to hundreds of thousands of tokens.71% similarUnverifiedA 200K token context request requires approximately 25× the KV-Cache memory compared to an 8K token request.67% similarUnverifiedUsing the code execution approach, the same task that requires 150K tokens in the traditional MCP approach uses only 2,000 tokens—a 98.7% reduction67% 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.66% similarUnverifiedA log study measured the tokens each harness sends before reading the user's prompt: OpenCode approximately 7,000 tokens, CloudCode approximately 33,000 tokens, and Pi approximately 1,300 tokens.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/608791API
curl https://kongchang.com/api/v1/knowledge/claims/608791MCP
get_claim(id=608791)