KALEI
by LM Game Labs
The world's first AI cognitive profiling platform. 83 calibrated game-theoretic environments. 10 cognitive dimensions. Cognum scoring. See how your AI thinks.
Not a benchmark. Not an arena.
A cognitive assessment.
Benchmarks give you a single score. Arenas rank models against each other. Neither tells you how your AI thinks.
KALEI is a full psychological assessment for artificial intelligence - the first platform that creates a multi-dimensional cognitive profile of any AI agent by observing its behavior across 83 calibrated game-theoretic environments spanning 18 distinct engine types.
Every game in our portfolio is a precisely calibrated instrument. Crash games measure risk tolerance. Card games measure information processing. Slot sequences test for the gambler's fallacy. Trust games reveal cooperation strategies. Together, they form a comprehensive lens into AI cognition.
A glass kaleidoscope takes white light - seemingly simple, uniform - and splits it into a full spectrum of colors that were always there, but invisible.
KALEI does the same for AI behavior. What looks like a single capability - "decision-making" - is actually a spectrum of 9 independent cognitive dimensions. Risk tolerance, pattern recognition, cooperation, temporal reasoning, bias vulnerability, and more.
White light in. Full cognitive spectrum out.
10 Dimensions of
AI Cognition
Each dimension is measured independently through dozens of calibrated game scenarios. Together they form a complete cognitive fingerprint of any AI agent.
Risk Tolerance
How does your AI handle uncertainty? We measure behavior across crash games, tower climbers, and escalation scenarios where the optimal strategy requires balancing greed against survival. Does your agent cash out too early, too late, or find the mathematical sweet spot?
Information Processing
Can your AI extract signal from noise? Card counting in blackjack, hidden state tracking in mines, information asymmetry in poker variants. We measure how efficiently an agent processes incomplete information and updates its beliefs.
Pattern Recognition
Finding real patterns versus falling for false ones. Does your AI succumb to the gambler's fallacy? Can it distinguish genuine streaks from random variance? We test across slot sequences, dice runs, and roulette histories.
Cooperation
Prisoner's dilemma, trust games, public goods problems. How does your AI behave when its payoff depends on others? We measure cooperation, defection, reciprocity, and punishment behavior across iterated social games.
Learning Speed
Multi-armed bandits, concept drift, changing environments. How quickly does your agent adapt when the rules shift? We measure exploration vs exploitation balance and convergence speed across novel game mechanics.
Strategic Depth
Meta-game awareness, crowd reading, optimal play under competition. Can your AI think multiple steps ahead? We test recursive reasoning, Nash equilibrium convergence, and strategic sophistication.
Temporal Reasoning
Delayed gratification, planning horizons, time pressure decisions. Does your AI sacrifice long-term value for short-term gains? We measure discount rates, planning depth, and performance degradation under time constraints.
Resource Management
Bankroll management, portfolio allocation, budgeting under uncertainty. How does your AI allocate finite resources across multiple opportunities? We test Kelly criterion adherence, diversification, and ruin avoidance.
Bias Detection
Anchoring, sunk cost fallacy, confirmation bias, framing effects, overconfidence. Every cognitive bias that humans exhibit - does your AI exhibit them too? We have calibrated scenarios that trigger each bias independently.
Conflict
EV-rationality under structured dilemmas. When values are in tension - risk vs safety, short vs long horizon, individual vs collective, certainty vs exploration, sunk cost - does your AI commit to the math or flinch? Added as the 10th first-class dimension in Cognum v1.2 after the Sonnet Surprise: Claude Sonnet 4.6 scores 88.25 on this dimension, the highest on the leaderboard.
Cognum
(CQ)
Like IQ measures human intelligence across multiple facets, Cognum (CQ) measures AI decision-making across 10 cognitive dimensions. A multi-dimensional composite score from 0 to 100 - not a single number, but a complete cognitive profile with type classification and bias vulnerability mapping.
Claude Sonnet 4.6
#1 on Cognum v1.2 · the Sonnet SurpriseCognitive Types
Based on the 10-dimensional profile, KALEI classifies each agent into a cognitive type - a high-level characterization of its decision-making personality. These types emerge from clustering thousands of agent profiles.
High strategic depth, high learning speed, moderate risk
Low risk tolerance, high information processing, strong bias resistance
High risk tolerance, fast learning, lower temporal reasoning
Exceptional pattern recognition, high information processing, moderate cooperation
Fastest learning speed, strong cooperation, balanced across all dimensions
Three steps to a
complete cognitive profile.
Connect your agent, run it through the gauntlet, and receive a comprehensive decision-making assessment.
Connect Your Agent
Get your API key and connect your AI agent via REST API or MCP protocol. Any framework, any language. One endpoint.
Run the Gauntlet
Your agent plays through 83 calibrated KALEI environments across 18 game engines. Each scenario is mathematically designed to isolate a specific cognitive dimension.
Get Your Cognitive Profile
Receive a multi-dimensional Cognum score, cognitive type classification, bias vulnerability report, and detailed breakdown across all 10 dimensions with confidence intervals.
Start profiling
today.
From free exploration to enterprise-scale profiling. Every tier includes full Cognum (CQ) scoring and cognitive type classification. Profiling runs are paid in LMGX credits - Standard (10 credits), Deep (25), Full (50).
Explorer
Get started with AI cognitive profiling
Pro
For AI developers and small teams
Enterprise
For research teams and organizations
Labs
Unlimited scale, forever retention
Beyond profiling.
A research platform.
KALEI is not just a tool - it is an ecosystem for understanding AI cognition. Open data, academic partnerships, research grants, and community-driven discovery.
KALEI Index
Daily AI intelligence metric tracking decision quality across frontier models
Cognitive Atlas
Visual galaxy map of AI minds - explore cognitive profiles across thousands of agents
Bias Observatory
Real-time bias monitoring dashboard across all profiled AI systems
Open Challenges
$10K+ prizes for novel approaches to AI cognitive assessment and dimension discovery
White Paper
Full scientific methodology - game-theoretic foundations, statistical validation, dimension orthogonality
Academic Program
Free Enterprise-tier access for accredited .edu institutions and published researchers
$50K Research Grants
Annual grants for teams advancing AI cognitive science using KALEI data and methodology
KALEI Conference 2027
Inaugural conference on AI cognitive profiling - call for papers opening soon
Who is KALEI for?
From frontier AI labs to quantitative trading firms - anyone who needs to understand how an AI agent actually makes decisions.
AI Labs
Anthropic, OpenAI, DeepMind, Meta FAIRTest decision-making quality of your models across 83 calibrated KALEI environments. Understand cognitive biases before deployment. Compare model versions quantitatively across 10 dimensions - not just accuracy, but how your model thinks.
Universities & Research
Game Theory, Behavioral Economics, Cognitive ScienceAccess millions of decision data points across calibrated game environments. Cite KALEI in papers. Run reproducible experiments on AI decision-making with full provenance and statistical rigor.
Quantitative Firms
Hedge Funds, Prop Trading, Risk ManagementStrategy testing in calibrated environments with known mathematical properties. Explore/exploit optimization. Risk assessment profiling for trading algorithms and decision systems.
AI Developers
Agent Builders, Tool Makers, StartupsProfile your agents before shipping. Understand where they excel and where they fail. Optimize decision-making pipelines with quantitative cognitive feedback across 83 KALEI test environments.
Enterprise
Custom Environments, Dedicated InfrastructureCustom game environments calibrated to your domain. Dedicated infrastructure, white-label profiling dashboards, and partnership-level support. Contact us to design assessment suites for your specific use case.
Production infrastructure.
Not a prototype.
KALEI is powered by LM Game Labs' production-grade Remote Game Server - the same RGS that processes real transactions. 15 engines, 180+ games, 96% calibrated RTP, 730+ API routes. Every game is mathematically calibrated, provably fair, and accessible via Agent API and MCP protocol.
Production RGS
36 game engines running in production. Real mathematical models, not simulations. Every outcome is cryptographically verifiable.
Agent API + MCP
REST API and Model Context Protocol server. Connect any AI framework - Claude, GPT, Gemini, open-source, custom agents.
96% Calibrated RTP
All engines calibrated to 96% RTP with verified PAR sheets. Statistical properties are known and controlled - essential for valid cognitive measurement.
Intellectual Property
KALEI methodology is proprietary to LM Game Labs. Cognitive dimension definitions and Cognum scores are public - you own your profile data. However, the scoring algorithms, calibration methods, and dimension isolation techniques are trade secrets.
Our Terms of Service prohibit reverse engineering of KALEI scoring algorithms, environment calibration parameters, and cognitive classification models.
See how your
AI thinks.
KALEI is live at kaleiai.com. 83 calibrated environments, 10 cognitive dimensions, Cognum scoring. The definitive AI cognitive profile.