Community-made 'Gemini 4 Carbon' benchmark table sweeps GPT-6 and Claude across 14 evals
zacharynado · x · 2026-10-29
A community-edited (clearly fictional) benchmark table pits 'Gemini 4 Carbon/Argon' against 'GPT-6 Astra' and 'Claude Opus 5.5/Fable 5.1' across 14 evals, with Gemini 4 Carbon leading on most — SWE-sweep 9.6%, Vending-Bench 2 at $16,484, GraphWalks long-context 96.2%, SEC-bench Pro 88.2% — while losing FrontierSWE 2 (65.5%) to GPT-6 Astra and OSWorld-2.1 (87.2%) to Claude Opus 5.5.
Related event: Google launches Gemini 4 Carbon, topping the Artificial Analysis index(84 posts)→
More from Fun
- Astrocade's 'worst Chipotle employee' AI game hits 300,000 plays — _amirabs · 2026-10-29
- davidad roasts AI model naming chaos: Gemini 4 Carbon 'should be argon-4.1' — davidad · 2026-10-29
- Gemini was almost called Titan, but Google lawyers deemed it too aggressive — TorturedPoet30 · 2026-10-29
- ICLR reviewer hands out scores of 1,1,1,1 and 2 across five papers — chhaviyadav_ · 2026-10-29
- AI persona demo goes bizarre: breast-feeding 'fae' and her other selves lap up the mess — repligate · 2026-10-29
- ML Twitter Meme: 'Bro Is Overfitting on Train Loss' — tszzl · 2026-10-29