DeepSeek 1017 accused of wild benchmaxxing: great scores, not a frontier feel
teortaxesTex · x · 2026-11-08
A viral critique argues DeepSeek 1017 was "wildly benchmaxxed": strong on benchmarks but it "doesn't feel anything like a frontier model," not even an older one like 5.6 Sol. The poster adds that people claiming parity with frontier models were just justifying their hardware purchases. The quip "finally a benchmaxxing benchmark" captures the community's skepticism over benchmark-gaming culture.
More from Models
- Muse token pricing: 2B tokens for $35, plus 1B-token referral code — RichardsonDx · 2026-11-08
- Same model, same config: Reddit user documents rare episodes where agents act like a lousy support rep — No_Dragonfruit_1599 · 2026-11-08
- Unverified report: DeepSeek V4 Pro quietly updated, hard agent task pass rate jumps 16 points — teortaxesTex · 2026-11-08
- User frustrated as Anthropic monthly promo credits expire within 30 days — arthurcolle · 2026-11-08
- Dev calls out Anthropic hype: Boris Cherny claimed coding was 'solved' in February amid buggy Claude Code — kuza55 · 2026-11-08
- MineBench quirk: cheaper models Sonnet 5.5 and Sol 6.1 both attempt knight shadows, Opus doesn't — Separate_Tonight9533 · 2026-11-08