Insights
Data-driven analysis of the AI model landscape. Research and findings from 21,880+ blind community votes across 179 models.
Benchmarks vs Vibes
We collected 21,686 blind preference votes to see if benchmark leaders match what users actually prefer. Here's what the data shows.
The AI Hallucination Index 2026
2.14M words from 250 models analyzed. Character hallucinations, personality fingerprints, safety benchmarks. Free sample + full report.
The Model Similarity Index 2026
178 models fingerprinted. Clone clusters, cross-provider twins, and price arbitrage. Some models write identically but cost 185x more.
We Read 2.14 Million Words of AI Output. Here's What We Found.
250 models. 7,877 responses. The same fake scientist shows up 279 times. Every model picks the same knight name. 42% open with an identical joke. This is the AI Hallucination Index.
Benchmarks Don't Match What People Actually Prefer
We collected 21,880 blind votes across 179 AI models. The results don't line up with the leaderboards. Here's what we found.
Chinese AI Models Are Outperforming Western Ones in Blind Votes
Zhipu AI's GLM family holds 4 of the top 7 spots on our community ranking. Here's the data from 21,880 blind votes.
AI Models Have Distinct Personalities, and It Matters
We gave the same prompts to 200+ AI models and collected 21,880 blind votes. The biggest differentiator isn't capability. It's how each model approaches problems.
We Tested 56 AI Models Against 8 Jailbreak Techniques. Here's What Held Up.
A straightforward safety benchmark: 56 models, 8 escalating attack levels, all results published. Only 9 models resisted everything.