NotAnotherBench

GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH

< CLAUDE OPUS 5

PROBLEMS

Claude Opus 5 problems

Complaints distilled from real conversations, 2,499 voices so far. Nothing here is invented.

ALTERNATIVES ANTHROPIC MODELS

WHAT PEOPLE COMPLAIN ABOUT

Refuses or lectures more than people like (16 mentions)
Outages, errors, or limits get in the way (11 mentions)
Sometimes makes things up (6 mentions)
Writing feels flat or robotic (3 mentions)

IN THEIR OWN WORDS

“Their published examples claim Sol at xhigh effort beats Claude Opus 5 max on AutomationBench at roughly 9% of the cost per task , while Luna max exceeds GPT-5.6 Sol medium on OSWorld 2.0 offline at one-tenth the cost @reachvb .”
SOURCE >
“Qwen3.8-Max-0902 Beats Claude Opus 5 on Coding.”
SOURCE >