Obscure-facts benchmark: DeepSeek V4.1 scores 33/33 vs MiniMax's 16/33, faster with far fewer tokens
teortaxesTex · x · 2026-10-20
User teortaxesTex ran a set of obscure-fact quizzes (twins of the hardest items) comparing DeepSeek V4.1-Flash-1017 against MiniMax M3.1-Flash: DeepSeek scored a perfect 33/33 vs MiniMax's 16/33, always faster and usually with far fewer tokens (818 tokens at 205 tok/s vs 10.8K tokens at 119 tok/s on one opera-history question).
- Sonnet had proposed a twist that DeepSeek's edge should shrink when names are replaced by descriptions or variant spellings. It failed: the gap grew (100% vs 50%), suggesting V4.1 has vastly more capacity.
- MiniMax's errors fit missing knowledge — invented second parties varying across runs (Hind/Cassini/John G. Adams), drifting dates and heights; DeepSeek was identical across runs.
- Feeding Sonnet MiniMax's CoT triggered anti-distillation/frontier research filters.
- Verdict: V4.1-Flash-1017 is markedly less hallucinatory, has vastly stronger knowledge, and is faster.
Related event: DeepSeek Aces Obscure-Facts Quiz, Crushes MiniMax in Head-to-Head Test(2 posts)→
More from Models
- Grok Bot DAU nearly triples in a month, from ~155K to ~450K — XFreeze · 2026-10-20
- Claude users report subscription limits running out much faster than before — prasenx · 2026-10-20
- Gemini users report extreme slowness lately, with Flash 3.8 hit hardest — cardrecommend03 · 2026-10-20
- User reports ChatGPT todos silently vanish between messages, threatens switch to Grok — koltregaskes · 2026-10-20
- doubleAI, co-founded by Mobileye CEO, claims its model cuts compute costs by 67% — pstAsiatech · 2026-10-20
- Longtime Anthropic critic returns: Claude 5.5 lineup is the best in months — intellectronica · 2026-10-20