This is the site to use when you want to decide “which AI model should we actually use?” with numbers rather than gut feeling. Artificial Analysis benchmarks AI models and the API providers that serve them from an independent position, lining up quality, response speed and pricing on the same footing. It is not limited to language models: it spans coding agents and image, video and speech generation models, and offers industry-specific indexes for fields such as finance and law. The public leaderboards can be viewed without signing in, and paid plans are available if you want to pull the data through an API. AI labs and major tech companies frequently cite it when talking about the performance of their own models, and it operates as an analysis firm the industry refers to.
Key Features
- Intelligence Index: A single score that combines multiple benchmarks into an overall measure of a model’s intelligence. The latest version as of September 2026 is v4.3, and the component evaluations are refreshed periodically
- Coding Agent Index: An index that measures how far coding agents can get on realistic, practical tasks. It is designed to look at the ability to finish work, not at how well a model converses
- Image / Video / Speech Arena: Leaderboards that compare image, video and speech generation models head to head. Generation quality is hard to quantify, so rankings are built up from comparative voting
- Capability Indexes: Indexes that isolate performance in specific domains such as finance and law. They reveal strengths and weaknesses that an overall score on a general-purpose benchmark hides
- Model Recommender: Enter the conditions you care about — whether you are prioritizing quality, price or speed — and it suggests models that fit those requirements
- Optima: A tool for building your own custom benchmarks. It lets you evaluate use cases that public metrics cannot measure against your own criteria
Pricing
| Plan | Price | What’s included |
|---|---|---|
| Public leaderboards | Free (no sign-up) | Viewing leaderboards, the various indexes and articles |
| Pro | $417 per seat per month (annual billing: $4,020 per seat per year, about $335 per month) | API access to the data, exports, custom charts and tables, industry reports, email support |
| Enterprise | Contact sales | Everything in Pro plus the highest API rate limits, compute market models and forecasts, workshops, AI strategy advisory and dedicated support |
Pricing is current as of September 2026. Check the official site for the latest information.
Enterprise is presented as a plan aimed at organizations with 150 or more employees. Note that there is no plan explicitly labeled as free: the split is between pages you can view without signing in and the paid Pro plan and above.
Pros & Cons
✅ Pros
- It shifts the model selection discussion from “this one seems good” to “here is the metric and here is the difference,” which makes it easier to reach internal agreement or explain a choice to a client
- Speed and pricing sit on the same screen as quality, so you can pick a model that balances budget against the responsiveness users actually feel
- Language, image, video, speech and coding agents can all be tracked in one place, saving you from hunting down a separate ranking for each field
- It is independent of any particular AI vendor and is cited by large companies including Microsoft, Meta, Google and Amazon, which makes it easy to point to as a source
⚠️ Cons
- Core features such as API access to the data and exports require the Pro plan at $417 per month or above. This is not a price point an individual signs up for casually
- Enterprise pricing is quote-based, so it is hard to get a sense of the cost early in an evaluation
- There was an episode in which the scoring of GPT-6 Astra drew questions about the validity of the evaluation (the scoring method was subsequently revised from Intelligence Index v4.2 onward)
- The benchmarks are ultimately its own designed metrics and do not guarantee performance on your actual use case
Comparison with Similar Services
| Service | How it differs from Artificial Analysis |
|---|---|
| LMArena (formerly LMSYS Chatbot Arena) | Built on Elo ratings from human votes, which makes it closer to a popularity vote than a benchmark score |
| Hugging Face Open LLM Leaderboard | Community-run and centered on academic benchmarks; it does not cover pricing or speed data |
| OpenRouter | Primarily a service for routing model APIs and handling billing; benchmarking is a secondary feature |
| Vellum AI LLM Leaderboard | A smaller, more commercially oriented leaderboard that covers a more limited set of models |
Who Is It For
- Developers and technology decision-makers choosing an LLM API for production while balancing cost against performance
- Creators and product owners who want to compare image, video and speech generation models across the board and find the right fit
- Enterprise AI adoption leads who want to narrow down candidate models using domain indexes close to their own field, such as finance or law
- Anyone who wants to embed citable benchmark data into an internal model selection report or dashboard
Summary
As the number of AI models keeps growing, a place where you can see trade-offs like “fast but expensive” or “cheap but slow” in a single view is genuinely valuable. Artificial Analysis is useful enough just to browse through the public leaderboards. A practical approach is to start by comparing your candidate models within the free range, and only weigh the cost of the Pro plan once you actually need to pull the data through an API or roll it into internal materials. Benchmark numbers are a starting point for a decision, so it is worth planning to test the finalists on your own tasks as part of the same process.