Scoring Transparency
Every ranking on LM Market Cap is computed algorithmically from a multi-source live data pipeline that aggregates upstream provider APIs, benchmark sources, and pricing feeds. Data is refreshed hourly. There is no manual intervention, no expert panel, and no editorial curation. The algorithm is the only thing that determines rankings.
How it works
A fully automated pipeline with no human decision points.
Hourly Data Fetch
A cron job pulls the latest model data every hour from multiple upstream provider APIs and benchmark sources, capturing pricing, context sizes, capabilities, and availability.
Score Computation
Each model is scored across six dimensions using a deterministic algorithm. The same input data always produces the same scores -- no randomness, no manual overrides.
Cache Update
Computed scores are written to a local JSON cache that the web application reads at runtime. Rankings update automatically with each refresh cycle.
Six Scoring Dimensions
Each model receives a composite score from 0 to 100, calculated as a weighted sum of six dimensions. The weights are fixed and published here.
Capabilities
25% weightWhat the model can do: coding, reasoning, instruction following, and multi-domain knowledge. Derived from benchmark data and capability flags reported by the provider.
Pricing Tier
25% weightCost-effectiveness based on input and output token pricing. Lower cost per quality point scores higher, making it easy to find the best value for your budget.
Context Window
15% weightMaximum context length the model supports. Larger context windows enable processing longer documents, codebases, and conversation histories.
Recency
15% weightHow recently the model was released or last updated. Newer models receive a freshness boost that decays over time, reflecting the fast pace of AI development.
Output Capacity
10% weightMaximum number of output tokens the model can generate in a single response. Higher output capacity matters for long-form generation tasks like code and articles.
Versatility
10% weightRange of modalities and tasks the model supports, such as text, code, function calling, and image understanding. More versatile models score higher.
What This Means In Practice
- Rankings are deterministic. Given the same input data, the same scores are always produced.
- No model provider can pay for a higher ranking. There is no sponsorship or promotion tier.
- Data refreshes every hour via cron. When a provider updates pricing or capabilities upstream, our scores reflect the change automatically.
- The scoring weights (25/25/15/15/10/10) are hardcoded. They do not change between runs.
- New models appear in rankings automatically as soon as they are published upstream. No manual approval is needed.
- All scoring logic runs server-side. There is no client-side computation that could be manipulated.