Browse the leaderboard

Browse the Stratix leaderboard — top models per capability and per benchmark.

The Stratix leaderboard surfaces top-performing models, organized by capability and by individual benchmark. It's the "who's winning" view of the public catalog.

How to read it

Two layouts:

• By capability — top N models in reasoning, code, math, multilingual, vision, multi-turn, tool-use
• By benchmark — top N models on a specific benchmark (MMLU, HumanEval, GSM8K, etc.)

What "top" means

The score on a single benchmark, ranked descending. The leaderboard isn't a holistic capability ranking (no benchmark covers everything). Use the leaderboard to find candidates; use Compare models to evaluate them against each other on the benchmarks closest to your task.

Refresh cadence

Scores update as the public evaluation corpus refreshes. New models appear when LayerLens (or a partner) publishes evaluations against the relevant benchmarks. Each leaderboard page shows a "last updated" timestamp.

Where to next

• Quarterly reports — narrative around leaderboard movement
• Models catalog
• Compare models
• Concept: Models and benchmarks