Claude Opus 5 beats Fable 5 on code — but max effort makes it worse
anthropic's claude opus 5 beats fable 5 on coding and knowledge benchmarks at half the price, but 'max effort' settings actually lower performance scores.
anthropic launched claude opus 5, offering near-fable 5 intelligence at half the token cost ($5 input/$25 output vs $10/$50). benchmarks show opus 5 leads gpt-5.6 sol and fable 5 on frontier-bench v0.1 coding (43.3%) and arc-agi-3 (30.2%), though it trails on deepswe and health tasks. a new fast mode increases speed 2.5x. unusual internal testing revealed the 'max effort' setting sometimes reduces performance scores while raising costs, suggesting users should stick to 'medium' or 'xhigh'.