Gemini 4 Neon hits 59% on rare-disease benchmark, near Claude Fable 5.2 at lower cost
danielmckinn0n · x · 2026-11-22
Developer Daniel McKinnon shares his hands-on experience with GamowLabs' RareBench rare-disease benchmark (41 configurations, 122 cases, metric: complete causal recall Rank@K):
- Gemini 4 Neon scores 59.0%, nearly matching Claude Fable 5.2 at 61.5% (Gemini 4 Carbon EAP also 61.5%, top), while being much cheaper with reasonable biological safeguards
- The author uses AI to help diagnose children with rare diseases; he previously could rarely use Fable end-to-end and often fell back to Opus 5
- He argues Google's new releases unlock a new paradigm of AI-assisted scientific research
Other scores: Gemini 4 Argon 57.4%, Claude Opus 5.5 51.6%, GPT-6.2 Astra 50.8%, Gemini 4 Hydrogen 49.2%.
More from Models
- mark_k: GPT-6.2 Sol is 'so freaking efficient', barely uses tokens — mark_k · 2026-11-22
- Yacine: Claude 5.5 'just not very good', pales compared to Neon — yacineMTB · 2026-11-22
- Wenhu Chen Revives 'But Can They Reason?' Debate, Asking Why Critics Went Silent — WenhuChen · 2026-11-22
- Tesla AI lead: new system uses less compute than the old one did — aelluswamy · 2026-11-22
- JUP-Eval tests whether models hold correct answers under user pushback, not just once — geoffwolfe · 2026-11-22
- Dev praises Gemini's vivid code comments: 'like an open kitchen drawer' — AI_Andrew · 2026-11-22