SWE-bench Verified (June 16): Fable 5 at 95.0%, Mythos 5 at 95.5%, Opus 4.8 at 88.6% — but Only 1 of 100 Scores Is Independently Verified

Digital Applied·medium signal

Current SWE-bench Verified standings put Claude Mythos 5 at 95.5%, Fable 5 at 95.0%, and Opus 4.8 at 88.6%. The critical caveat: of 100 models on llm-stats, only one carries an independent verification badge — the other 99 are vendor-submitted. The same model family scores 51.90% (Opus 4.6 on Scale's harness) vs 69.2% (Opus 4.8 on Anthropic's harness), a 17–21 point spread driven by scaffolding, not the model.