Opus 5 beats Fable on benchmarks but loses badly in real use, analyst says
petergyang · x · 2026-07-26
- The quoted analysis argues that current public benchmarks are becoming nearly useless for judging frontier models.
- It claims Opus 5 looks much weaker than Fable in real use, even though Opus wins on many benchmarks.
- The takeaway is that domain-specific evals built on private datasets may become more trustworthy than public leaderboards.
- It also suggests Anthropic may be training the 5-series differently from earlier Sonnet/Opus generations, though the post truncates before the full comparison.
More from Models
- Rumor claims Kimi faked performance by serving Claude; DeepSeek new model surprises in evals — realsohamparekh · 2026-09-11
- GPT-5.6 writes well but is instantly forgettable, user complains — BasedRaddka · 2026-09-11
- Opus Refuses Protein Research Codebase Over 'Safety' Concerns, Dev Considers Rolling His Own — josephdviviano · 2026-09-11
- User Hails Unconfirmed 'DeepSeek 4.1 Flash' as an Inflection Point in LLMs — himanshustwts · 2026-09-11
- Terminal Bench v4: GLM-5.3 Leads at 41.9%, Kimi-K3 Underwhelms at 12.6% — Ok_Warning2146 · 2026-09-11
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11