Research-math benchmark: DeepSeek V4.1 Flash scores 3.038, near-tied with MiniMax M3.1 Flash
An evaluator puts DeepSeek V4.1 Flash 1017 at 3.038, in a near-tie cluster with Meta Muse Wonder, Gemini 4 Argon, Mimo v2.6 Pro and MiniMax M3.1 Flash, strong at obstruction finding but weak on theorem scope.
3/ DeepSeek v4.1 Flash 1017 is a strong full-bench research-mathematics system and scores 3.038. It belongs to the near-tie cluster with Meta Muse Wonder, Gemini 4 Argon, Mimo v2.6 Pro, and MiniMax M3.1 Flash. Its defining strengths are decisive obstruction finding, broad theorem-level partial progress, good statement debugging, and complete structured coverage. Its principal weakness is theorem-scope control: several polished, highly confident claims rely on literature results that do not apply in the asserted generality. DeepSeek is therefore an excellent high-throughput conjecture debugger and research assistant, but literature-dependent solved claims still require strict external verification before acceptance.