MalwareTech: GLM 5.3 is my top price/performance pick, abliterated version is worse
MalwareTechBlog · x · 2026-11-09
Security researcher MalwareTech (Marcus Hutchins) confirmed in his spat with Matthew Green that he has substantively used GLM 5.3, calling it his go-to model for the best price/performance trade-off in vulnerability discovery and exploit testing.
Notable details: he tried abliterated (de-aligned) GLM but dropped it from his vulnerability pipeline because the abliterated version is a worse model than the non-abliterated one; and he argues frontier models still refuse requests weaker models don't, even with safeguards like CVP/Daybreak — his reason for going unrestricted.
More from Models
- Anthropic bets on healthcare and biology as Claude sole-authored paper fuels hidden-model speculation — coherence · 2026-11-09
- $300/month SuperGrok user slams degraded Grok: 'barely usable,' voice mode also worse — alexcovo_eth · 2026-11-09
- Leaked Google internal checkpoint 'Krypton' reportedly outperforms 'Carbon' — lyraxana · 2026-11-09
- Developer Jason Kneen kicks off hands-on testing of Qwen 4 — jasonkneen · 2026-11-09
- GLM matches Opus at mass-converting Windows drivers into LPE 0days, says security researcher — MalwareTechBlog · 2026-11-09
- Nous Research Arete-27B GGUF quants now available on Hugging Face — gladkos · 2026-11-09