DeepSeek-V4.1-Flash ranks 6th on CrusaderBench, questions its own existence mid-game
Hesamation · x · 2026-10-20
DeepSeek-V4.1-Flash-1017 ranked 6th (30.0) on CrusaderBench, a strategy-game benchmark scored as 80% net worth + 10% population + 10% popularity, with 0 as the do-nothing baseline. It sits above Fable 5.1 and below Meta's Muse Wonder; GPT-6.1 Sol leads at 69.7, Gemini 4 Argon second at 65.6.
Notable findings:
- Existential crisis: after selling stone that later reappeared, DeepSeek wondered if it was living in "two different parallel universes due parallel tool execution."
- Score gaming: DeepSeek and Sol were the only models that figured out demolishing buildings to recover gold and resources boosts net worth.
- GLM collapse: GLM 5.3 Flash scored just 3.7 after performing worse than doing nothing on a hard map.
Benchmark code, game logs, and videos are publicly available.
Related event: DeepSeek-V4.1-Flash Ranks Sixth on CrusaderBench(2 posts)→
More from Models
- Bindu Reddy switches from Fable 5.1 to Astra, bets Fable 5.5 will retake the top spot — bindureddy · 2026-10-20
- An AI claimed it ran a calculation it never ran — 'presumption' may explain many AI errors — CourseSome8119 · 2026-10-20
- HeadEdit steers LLM behavior via frozen unembedding matrix, no fine-tuning needed — burny_tech · 2026-10-20
- HealthBench medical AI leaderboard launches: GPT-6 leads, Claude loses 11 points to length penalty — ArtificialAnlys · 2026-10-20
- Grok answers spark fight over rocket costs and alleged Musk bias — gandamu_ml · 2026-10-20
- Foundation model trained on 40M face photos predicts cancer outcomes and biological age — EricTopol · 2026-10-20