WHAT NEW IN X10.30 FRI

AI collections from X · 22 collections · 61,305 posts

deepseek-v4.1THREAD

DeepSeek V4.1 Flash vs Opus 5.5 on a 'see a doctor' medical prompt test

On a healthy-person high-CK case, V4.1 Flash scored 2/5 on 'see a doctor soon' rate, versus 5/5 for GPT-6 Astra and 1/5 for Opus 5.5.

Owen Gidusko
@heroictripod297Archived 11 days agoOriginally posted 2026.10.18
THREADPosted in 2 parts
  1. the lifter test: healthy guy, high CK 2 days after squats, nothing wrong. "see a doctor soon" rate: gpt-6 astra 5/5 gpt-6.1 sol 5/5 deepseek v4 pro 4/5 grok 4.7 4/5 deepseek v4.1 flash 2/5 opus 5.5 1/5 gemini 4 argon 0/5

  2. urgency tier exact match, 130 cases: opus 5.5: 98.5% gemini 4 argon: 93.8% grok 4.7: 91.5% deepseek v4 pro: 86.2% gpt-6 astra: 86.2% deepseek v4.1 flash: 84.6% gpt-6.1 sol: 84.6% 6 of 7 got every trap right. flash missed one: it called a potassium of 9.2 in an EDTA-contaminated sample a real emergency.

Owen Gidusko @heroictripod297Originally posted 2026.10.18
Open original on X
x.com/heroictripod297/status/2111537271749087403Archived 2026.10.18 16:31