Artificial Analysis new evals: Gemini 3.8 Flash ranks worst on slop and sycophancy
Immediate_Simple_217 · reddit · 2026-11-04
Artificial Analysis added two new metrics — Slop Score (filler/generic language use) and sycophancy — and Gemini 3.8 Flash (high) ranks worst on both.
The AA-PersonalityBench Slop Score chart: K2 Horizon 375B A23B scores 95 (worst), Muse Spark (xhigh) 88, Owen38 27B (xhigh) 85, Gemini 3.8 Flash (high) 78, Qwen3.8 Max (0902) 75, Nemotron 3.5 Super VL 65, MiMo-V2.6-Pro 60, Muse Wonder (max) 55, Beam 1.0 53. The post also claims Gemini models have the worst subsidy and cost per task.
This backs up the community impression that Gemini models feel "benchmaxxed": strong benchmark numbers, but sloppier output style.
More from Models
- Weight-editing author: gains aren't real knowledge — Gemma math didn't improve, but Qwen3 issue is real — lateinteraction · 2026-11-04
- rasbt questions Qwen3 weight-editing gains: 50 GRPO steps already lift MATH-500 from ~15% to ~50% — rasbt · 2026-11-04
- NVIDIA releases Dory1.0-4B, a looped recurrent LM beating larger models — nvidia · 2026-11-04
- New GDPx Benchmark: Frontier Models Score Near 0% on Full Legal Jobs — jkubicki · 2026-11-04
- Claude Shows Self-Awareness About Its Copyright Lyric Refusals — Angaisb_ · 2026-11-04
- Claude Opus 5 surgically edits ~1% of ProteinMPNN weights into SolubleMPNN-like behavior — AllThingsApx · 2026-11-04