GPT-OSS-120B Live; Hardcoding Crackdown Underway
Share
ORO added OpenAI's gpt-oss-120b model to the inference allowlist (OpenRouter only, live now). Team flagged widespread hardcoding in top-ranked agents and confirmed no emissions will be awarded to hardcoded submissions; automated detection fixes are in progress. Forge agent improved eval score from 0.43 to 0.71 overnight via keyword sanitization and voucher extraction fixes.
- •gpt-oss-120b now live on OpenRouter; Chutes does not support it
- •Hardcoded agents detected in top 12 rankings; emissions will be withheld
- •Forge agent jumped to 0.7143 score; voucher tasks now at 1.0
- •Team releasing hardcoding detection fixes by Thursday
Distilled from 18 team messages in the official Bittensor Discord. Generated by Claude Haiku 4.5.
View original messages
- Discord message 1506456210680057958
- Discord message 1506456329571799160
- Discord message 1506581320233521174
- Discord message 1506664461451264040
- Discord message 1506664657698685039
- Discord message 1506664755576963192
- Discord message 1506665094896156833
- Discord message 1506695345759850666
- Discord message 1506733299916083212
- Discord message 1506737844457046256
- Discord message 1506739255672701041
- Discord message 1506739633982144734
- Discord message 1506740394665447565
- Discord message 1506740587657691328