DeepSeek V4.1-Flash-1017 is its least hallucinatory model yet, but regresses on knowledge and LCR
teortaxesTex · x · 2026-10-19
Benchmarks watcher teortaxesTex rounds up results for DeepSeek V4.1-Flash-1017 on Artificial Analysis:
- Hallucinations: it's the least hallucinatory DeepSeek model of 2026, a clear improvement over predecessors on the AA-Omniscience index, where most older DeepSeek models score negative.
- Trade-off: it regresses on knowledge versus its predecessor, and also on long-context reasoning (LCR).
- Value: on the intelligence-vs-cost Pareto chart, DeepSeek V4.1 (45) is Pareto-dominated by GPT-6.1 Sol (51, $0.39 per task) at its peak price point.
- The author caveats that Artificial Analysis has its own flaws, e.g. the broken CritPt benchmark, so rankings should be taken with a grain of salt.
Related event: DeepSeek V4.1-Flash hits lowest hallucination rate but regresses elsewhere(2 posts)→
More from Models
- DeepSeek-V4.1-Flash-1017 scores 10+ points lower on third-party Terminal bench tests — DigiDecode_ · 2026-10-19
- Open decision model D3 tops Jev Index, ships vLLM recipes from 0.6B to 27B — AccBalanced · 2026-10-19
- SenseTime co-founder: encoder-free native VLMs key step toward unified multimodal AI — liuziwei7 · 2026-10-19
- Users complain Gemini's Argon process runs far too long — juliuslautreamont · 2026-10-19
- OpenAI Launches New ChatGPT Voice Powered by GPT Live — OpenAI · 2026-10-19
- User complains Claude psychoanalyzes you instead of giving direct answers — haider1 · 2026-10-19