Grok 4.7 is now the LEAST sycophantic model evaluated by Artificial Analysis, and the improvement across recent Grok releases is absolutely incredible
Artificial Analysis' AA-PersonalityBench shows that with every new release, Grok is becoming less sycophantic, more sincere, and producing significantly less unnecessary filler
Here's how Grok has evolved:
Slop (lower is better):
• Grok 4.5 — 68
• Grok 4.6 — 38
• Grok 4.7 — 6
Sycophantic (lower is better):
• Grok 4.5 — 18
• Grok 4.6 — 3
• Grok 4.7 — 1
Sincere (higher is better):
• Grok 4.5 — 26
• Grok 4.6 — 63
• Grok 4.7 — 90
That's a 62-point drop in Slop, a 17-point drop in Sycophantic, and a massive 64-point increase in Sincere across just three releases
Less unnecessary filler, less agreeing with everything users say, and more authentic, direct responses
Grok is becoming the kind of SI assistant that gives you straight answers instead of simply telling you what you want to hear. This is exactly the kinda superintelligence you need
Artificial Analysis' AA-PersonalityBench shows that with every new release, Grok is becoming less sycophantic, more sincere, and producing significantly less unnecessary filler
Here's how Grok has evolved:
Slop (lower is better):
• Grok 4.5 — 68
• Grok 4.6 — 38
• Grok 4.7 — 6
Sycophantic (lower is better):
• Grok 4.5 — 18
• Grok 4.6 — 3
• Grok 4.7 — 1
Sincere (higher is better):
• Grok 4.5 — 26
• Grok 4.6 — 63
• Grok 4.7 — 90
That's a 62-point drop in Slop, a 17-point drop in Sycophantic, and a massive 64-point increase in Sincere across just three releases
Less unnecessary filler, less agreeing with everything users say, and more authentic, direct responses
Grok is becoming the kind of SI assistant that gives you straight answers instead of simply telling you what you want to hear. This is exactly the kinda superintelligence you need
