Artificial Analysis V5 index shifts to harder benchmarks; reconstructed difficulty scale puts GPT-4 near zero
scaling01 · x · 2026-10-29
Artificial Analysis is now using substantially harder benchmarks in its Intelligence Index V5. A community-reconstructed chart maps each benchmark onto AA's Progress Index difficulty scale: MLCR, GDP.pdf (v5), FrontierCode 2.0 (v5), CritPt v1, and HLE-Diamond top the hardest tier, while AIME 2025, LiveCodeBench, MMMU-Pro, and GPQA Diamond are now among the easiest — and GPT-4 (2024.11) scores near zero. The same author previously noted OpenAI's clear RL edge, with models struggling sharply once reasoning is removed.
More from Models
- Google DeepMind unveils Gemini 4 Carbon, SOTA on cybersecurity and long-horizon tasks — rseroter · 2026-10-29
- Google launches Gemini 4 Carbon two weeks after Argon, tops AA Index at 59 and SEC-bench Pro at 88.2% — rseroter · 2026-10-29
- LoganK: Gemini 4 Carbon is a big jump over Argon, with better coding usability and alignment — legit_api · 2026-10-29
- Qwen Flash turns a sci-fi prompt into a full rocketry derivation before writing the story — LankyGuitar6528 · 2026-10-29
- Scaling cross-situational word learning: 2,441 hours of child-perspective video from Stanford team — RichmanRonald · 2026-10-29
- How this user kept Claude accounts unbanned: domain email routing + self-hosted IP — vista8 · 2026-10-29