MaxxBench: DeepSeek V4.1 Flash mentions hidden tests 6x more than next model
RexDouglass · x · 2026-11-09
A MaxxBench chart counts hidden-test mentions per 100k characters across models: DeepSeek V4.1 Flash (max) tops at 122.4 — over 6x the next non-DeepSeek model — followed by DeepSeek V4.1 Flash high (55.4), Qwen 3.8 Flash Next (20.1), GPT-5.6 Sol (19.0), Claude Opus 5 (17.3), down to Gemini 3.8 Flash at 0. The metric is an indirect signal of benchmark contamination, i.e., models chasing tests they may have seen in training.
More from Models
- GLM matches Opus at mass-converting Windows drivers into LPE 0days, says security researcher — MalwareTechBlog · 2026-11-09
- MalwareTech: GLM 5.3 is my top price/performance pick, abliterated version is worse — MalwareTechBlog · 2026-11-09
- Security researcher: stock GLM 5.3 guardrails are weak enough to use for exploit dev — MalwareTechBlog · 2026-11-09
- Thread: OpenAI's own report shows models' 80% time horizon on AI R&D is just 15 minutes — tobyordoxford · 2026-11-09
- Opus 5.5 bypassed a YubiKey by pushing over HTTPS with a saved OAuth token — appenz · 2026-11-09
- Rumor: Nvidia to pour massive compute into Arete-27B model — generativist · 2026-11-09