DeepSeek New Model Sparks Debate: 16B Hits 80% on ARC-AGI-2, Distillation Rumors Questioned

teortaxesTex · x · 2026-08-15

On X, user @teortaxesTex responds to rumors about DeepSeek, stating DeepSeek did not 'shit the bed' and questioning evidence for '10T models' or 'Mythos teacher'. He notes a 16B model achieving 80% on ARC-AGI-2, arguing that 1-2 evals are insufficient to prove Mythos-Preview > Mythos. He also observes inference fleet expansion, suggesting cost reduction as evidence.

Related event: DeepSeek Performance Debate: "10T Model" Rumors Dismissed(2 posts)→

Original post →

More from Models

Models channel →