Skip to content
The Next Token
News
How it works
Research
Products
Notes
About
RSS
Section
Research
Papers and ideas worth your time, read closely and explained without assuming a PhD.
Writing
1 piece
No. 012
Oct 22, 2026
23 min read
Coding benchmarks that leak their answers, and which scores to trust
Vals.ai found the fix still readable in two-thirds of the coding tasks Xiaomi released for MiMo-V2.6, and MiMo found it. We pulled 92 task images from SWE-bench Verified, SWE-bench Pro and Xiaomi's release to see which leaks are closed as of 20 October 2026, and collected what happens to scores when they are.
→