Dev says AA index lost credibility the day Opus 5 outscored Fable 5 on paper
HarveenChadha · x · 2026-09-05
Harveen Chadha argues that eval scores are increasingly unrepresentative of real-world usage: he stopped trusting the Artificial Analysis index the day Opus 5 outscored Fable 5, and finds it worse still that Muse Spark 1.3 beats Astra on the leaderboard. His takeaway: rankings and lived experience are diverging — don't pick models by benchmark scores alone.
Related event: Users Report Real-World Results Contradict AA Index Rankings(5 posts)→
More from Models
- Opus 5.5 tops OpenRouter spending in 20 of 29 task categories — mmurph · 2026-10-21
- Meta team builds open-source 30B LLM from scratch in months, hails 'second era' — TimDarcet · 2026-10-21
- DFlash drafter for Muse Glimmer-30B: 5 layers, blockwise 16-token spec decoding — TimDarcet · 2026-10-21
- EvoPress quantization shrinks 30B model from 56GB to 19GB at near-bf16 accuracy — TimDarcet · 2026-10-21
- Radar chart: Muse Glimmer-30B benchmarked against Gemma4-31B and Qwen3.6-27B — TimDarcet · 2026-10-21
- Privacy evals: Muse Glimmer-30B leaks less than Qwen3.6-27B on CIMemories — TimDarcet · 2026-10-21