Most realistic benchmarks?
retained text
AA and other benchmarks don't reflect what models can really do. Some benchmaxxed models are unusable for anything but poems, albeit being close to the frontier on coding benchmarks. What are the best, most realistic benchmarks you know for coding, agentic research, office work?
document fetched 2026-10-26 00:38 · canonical https://reddit.com/r/LocalLLaMA/comments/1xevwim/most_realistic_benchmarks