10K-run test: LLM hits 99.91% on 30-digit multiplication, 99.8% on division
srchvrs · x · 2026-09-23
srchvrs reports a scaled-up LLM arithmetic benchmark: 10K tests per operation found 30-digit multiplication 99.91% accurate, while 36-digit dividend / 6-digit divisor division scored 99.81% on quotient and 99.66% on remainder. He considers the results stable and plans to raise the test ceiling from 20 to 40 digits next.
More from Models
- Researcher's early take: opus 5.5 best since 4.5, but 5.6-sol still the best all-around model this year — DimitrisPapail · 2026-09-23
- Dev finds Opus 5.5 'very good', but 5.6-sol still the best all-around model this year — DimitrisPapail · 2026-09-23
- "Don't trust benchmarks" is no excuse when 99% of them show zero improvement or regression — Angaisb_ · 2026-09-23
- Gemini 3.8 Flash TTS priced at $32.98/1M chars, pricier than its predecessor — ArtificialAnlys · 2026-09-23
- 214 API calls later: Ling-3.0-flash-VL spots a 30-yuan receipt error and shines on video detail — alifcoder · 2026-09-23
- Audio8 ASR Infinite open-sourced: ultra-low latency 24/7 streaming ASR with no drift — TheMoonMidas · 2026-09-23