NEW LLM BENCHMARKS: The newest figures are in following the releases of Gemini 4 Neon, GPT-6.2 Astra, and GPT-6.2 Sol!
🏆 Who leads where:
• Claude Fable 5.2 claims #1 overall on both independent indexes—Artificial Analysis (65) and Vals Index (71.31%)—plus top spots in FrontierCode 2.0 (49.1%), Vibe Code Bench, and Arena WebDev (1,824 Elo).
• Gemini 4 Neon dominates the most domain-specific rows overall, sweeping Vals Finance (68.31%), Legal (78.36%), Tax Agent Bench (54.89%), and SWE-sweep (12.8).
• GPT-6.2 Astra holds the crown in science & reasoning—topping Vals Science (68.44%), Terminal-Bench-Science (77.1%), Agents' Last Exam (78.3%), and AA-3D (1,318 Elo).
Full independent vs. vendor-published breakdown below 👇
i genuinely believe this is the greatest moment in comedy movie history. i rewatch this movie all the time, knowing exactly what's coming, and somehow this scene still makes me laugh every single time
@theo If true, my guess is they just don't need it yet. Private money is still pouring in, and going public means quarterly earnings pressure on a company spending billions on training runs. Hard to build frontier models on a 3-month report card.
@bindureddy Two of these are really one prediction. When AI gets cheaper to use, people don't use the same amount for less money. They use far more. That's how GPU demand keeps rising while token prices fall.