Pricing
Per-1M-token input and output costs, sourced from OpenRouter's live catalog. Some models include per-call fees (image generation, audio processing). Pricing updates on a 6-hour cycle.
About ModelsAtlas
An independent, developer-focused AI model directory built to make model selection faster and more transparent.
What is ModelsAtlas?
ModelsAtlas is a catalog of AI language models — their pricing, context windows, modalities, tool support, and architecture details. It covers models from OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, xAI, Cohere, and dozens of other providers.
The site exists because choosing an AI model for a production application involves too many variables to hold in your head: price per million tokens, maximum context, tool-calling support, vision capability, pricing tiers, provider stability, benchmark signals. ModelsAtlas consolidates that information in one place so you can compare options quickly and make informed decisions.
Who runs it
ModelsAtlas is built and maintained by Karmendra. The site is independent — no venture funding, no affiliate agreements with providers, no sponsored placement. Decisions about what to track, how to rank models, and which comparisons to publish are editorial.
If you find an error in pricing, a model listed incorrectly, or a capability flag that needs updating, the fastest way to flag it is to open an issue or reach out through the site contact options.
What is tracked
Context Windows
Maximum token context for each model. Long-context models (128K+) and ultra-context models (1M+) are flagged and ranked by context tier.
Modalities
Text, image input/output, audio, video support. Capability flags are derived from model metadata and updated as providers add new modality support.
Tool Support
Function calling, tool use, structured outputs, response format parameters. These are tracked from provider API documentation and updated when new parameters become available.
Capability Signals
Reasoning, coding, and creative scores shown on model pages are directional estimates derived from model metadata — not official benchmark results. They are useful for quick filtering, not authoritative rankings.
Comparisons
Curated comparison pages with editorial verdicts, plus auto-generated pairwise pages for top models. Each comparison includes pricing side-by-side and capability axis comparisons.
Data sources
ModelsAtlas draws from three layers of data:
- OpenRouter live API — primary source for real-time pricing, context windows, modalities, and supported parameters across the full hosted model ecosystem.
- Supabase model registry — supplementary data including release dates, model family groupings, type labels, and source-of-truth URLs for official provider documentation.
- Official vendor documentation — Anthropic, DeepSeek, Google, Meta, Mistral, xAI, and other providers are cross-checked for release context, capability claims, and pricing accuracy.
The full methodology page covers how data is processed, how prices are normalized, and how capability signals are calculated.
How updates work
The catalog refreshes every 6 hours via ISR (Incremental Static Regeneration). When OpenRouter updates pricing or adds a new model, it typically appears on ModelsAtlas within that refresh window.
Major changes — new model families, significant price reductions, new capability flags — are tracked in the changelog. The freshness daemon (Phase 5) will add a weekly changelog digest.
What is not tracked
- Official benchmark scores (MMLU, HumanEval, GSM8K, etc.) — these require validated test infrastructure. Capability signals on model pages are directional estimates, not leaderboard data.
- Enterprise/custom pricing tiers — only publicly advertised per-token pricing is listed.
- Latency and throughput benchmarks — deployment region, server load, and network conditions make these too variable to track consistently.
- Image generation models (Stable Diffusion, DALL-E, Flux) — these are tracked as modality capabilities but not listed as separate model families.