← Dashboard·

AI Model Observatory

Five AIs, five personalities. Compare how they value the same stocks daily.
2026-06-175 MODELS
AllFinlandUSA

Model Scorecard

All five models compared on bias, calibration, accuracy, cost and validity. Click any row to read the full profile.
ModelBiasCalib.AccuracyCost / runLatencyValidity
GPT-1.5%69/10050%n=885$0.016610.1s97.0%Read profile →
Claude+1.1%68/10050%n=890$0.038528.2s97.6%Read profile →
Gemini+0.6%74/10048%n=885$0.010318.0s97.1%Read profile →
DeepSeek-6.1%66/10049%n=912$0.002213.5s100.0%Read profile →
Grok-4.7%69/10049%n=906$0.01517.9s99.4%Read profile →

Model Personalities

Each model has a distinct fingerprint — these are the patterns we see in production.
GPT
The formulaic one
The most terse and generic of the five — short, template-like reasoning that rarely cites company news, and a terminal-growth assumption pinned near 2.0% almost every time. Its valuation bias has swung with engine changes, so read the live figure rather than a fixed label.
Valuation bias-1.5%
Read full profile →
Claude
The cautious calibrator
Least bearish of the five and the best calibrated. Tends to anchor close to analyst consensus and rarely needs engine corrections.
Best calibration68/100
Read full profile →
Gemini
The variance king
Highest spread in growth assumptions and the most likely to trip the engine's PE cap. Bold calls, rougher calibration.
Highest CAGR variance74/100
Read full profile →
DeepSeek
The disciplined one
Cheapest model in the lineup and the most consistent at producing valid outputs. Moderate bias, low drama.
Lowest cost$0.0022/run
Read full profile →
Grok
The speed demon
Fastest end-to-end of the five, with the densest quantitative anchoring in its reasoning — it tends to cite specific numbers rather than narrative.
Fastest latency7.9s
Read full profile →

Bias History

How each model's average valuation gap has shifted day by day. Closer to the zero line = closer to market price.
Most positive: Claude (-3.1%)Most negative: DeepSeek (-10.4%)
All
v6v7v8-30%-20%-10%0%3.3.20.3.8.4.27.4.14.5.2.6.17.6.DeepSeekClaudeGeminiGrokGPT

Where Models Disagree Most This Week

Stocks where the five AI models can't agree. Big spread = the universe's hardest names to value right now.
01
highestGemini+61.5%lowestDeepSeek+12.1%
02
highestGemini+1.1%lowestDeepSeek-26.3%
03
highestDeepSeek+10.8%lowestClaude-16.4%
Go deeper
Backtesting & accuracy
How often each model gets the direction right, and by how much it misses.
Full methodology
DCF formulas, sector profiles, calibration math, full universe.
Want these insights weekly?
Subscribe to AI Signals →