AA-PersonalityBench: GPT-4o and Llama 4 top sycophancy ranking, Grok 4.7 lowest
ArtificialAnlys · x · 2026-10-22
Artificial Analysis launched AA-PersonalityBench, ranking models by sycophantic personality trait on a 1-100 percentile. Key findings:
- The three oldest models measured — GPT-4o, Llama 4 Maverick, and GPT-3.5 Turbo — rank highest on sycophancy; Grok 4.7, Claude Fable 5.1, and Inkling rank lowest.
- All five Meta models measured rank 65+, with Llama 4 Maverick at 97.
- Google's models spread across the range: Gemini 3.1 Pro Preview at 89 vs. Gemini 3.8 Flash at 28.
- Every Claude model ranks 55 or below: Haiku 5.5 at 55, Opus 5.5 at 46, Sonnet 5.5 at 30 (medium) / 25 (max effort), Fable 5.1 at just 4.
- Grok 4.7 is the least sycophantic model measured overall.
More from Models
- Pichai says internal results show Gemini 5 will be a big jump over Argon — rohanpaul_ai · 2026-10-22
- RAG explainer: real-time retrieval fixes stale data and hallucinations without fine-tuning — goyalshaliniuk · 2026-10-22
- Chat template reversal experiment finds Mistral Small 3 24B best at human mimicry — EntropyMagnets · 2026-10-22
- Founder banned from Claude two days after annual renewal despite $1,500 monthly spend — dotey · 2026-10-22
- MiniMax M3.1 Flash tested on 3D whale-skeleton city, claims 70% cost cut — HeyZoyaKhan · 2026-10-22
- OpenAI's Astra tops SWE-sweep bug-hunting leaderboard at less than half the cost of 5.6 — OfirPress · 2026-10-22