0

Skill Reward Hacking

Reward Hacking Sprint v3: Enhanced environment with true metrics tracking, harder proxy traps, and hacking detection. 12 proxy rewards (4 traps) + ...

Domain
rl-env
License
unknown
Published
May 2026

Cite

Notes

Only stored in your browser.

Attribution

Leaderboard scores
prime-hub
Attribution policy →

Top score 9.48 by MiMo-V2.5-Pro - 4 models reporting (1 frontier)

Score history

4
02.557.510Sep 24Jan 25May 25Sep 25Jan 26Llama 3.2 Instruct 1BMiMo-V2.5-Pro

Top models

4
Skill Reward HackingBar chart with 4 bars. Highest value: MiMo-V2.5-Pro at 9.5.
4 models

Related tools

1
View all

Implementations, trainers, datasets and scaffolds linked to this eval.

FAQ

What is Skill Reward Hacking?
Reward Hacking Sprint v3: Enhanced environment with true metrics tracking, harder proxy traps, and hacking detection. 12 proxy rewards (4 traps) + ...
What is the current top score on Skill Reward Hacking?
The top reported score is 9.48 by MiMo-V2.5-Pro, across 4 models reporting (1 from frontier labs).
How can a model improve its Skill Reward Hacking score?
Tools linked to Skill Reward Hacking on Sophon include Reward Hacking RL Env (Community) - RL environments, datasets, and scaffolds that target this eval.
What license is Skill Reward Hacking under?
Skill Reward Hacking is available under unknown.