Loading…
Loading…
Measured cost and capability, side by side. Compare exact model configurations on the same benchmark.
Intelligence Index v5
| Model / configuration | USD / task | index points | Compare |
|---|---|---|---|
Cohere | $0.00 | 6.2v5 | |
Institute of Foundation Models | $0.00 | 17.8v5 | |
Cohere | $0.00 | 3.4v5 | |
OpenAI | $0.0051 | 9.7v5 | |
Mistral | $0.0399 | 4.3v5 | |
IBM | $0.0405 | 5.1v5 | |
OpenAI | $0.0422 | 18.5v5 | |
Anthropic | $0.0438 | 18.6v5 | |
OpenAI | $0.0713 | 29.4v5 | |
OpenAI | $0.0714 | 9.4v5 | |
OpenAI | $0.0729 | 21.4v5 | |
Anthropic | $0.0918 | 24.6v5 | |
OpenAI | $0.1041 | 23.5v5 | |
Meta | $0.1162 | 34.1v5 | |
Xiaomi | $0.1285 | 26.7v5 | |
Meta | $0.1359 | 36v5 | |
OpenAI | $0.152 | 27.9v5 | |
Anthropic | $0.1586 | 29.9v5 | |
OpenAI | $0.1615 | 34.8v5 | |
OpenAI | $0.1688 | 38.2v5 | |
Meta | $0.1695 | 6.3v5 | |
Google | $0.1945 | 6.2v5 | |
Thinking Machines | $0.2032 | 12.9v5 | |
OpenAI | $0.2259 | 30.5v5 | |
Anthropic | $0.2484 | 34.6v5 | |
Xiaomi | $0.2572 | 29.9v5 | |
OpenAI | $0.2593 | 42.4v5 | |
Reflection | $0.2844 | 17.7v5 | |
NVIDIA | $0.3145 | 5.5v5 | |
OpenAI | $0.3313 | 23.8v5 |
Cost per task is Artificial Analysis’s measured average across the Intelligence Index evaluations, including input, cache, reasoning and answer token costs. It is not a token price, a monthly estimate, or the cost of every task succeeding. Each row is an exact model configuration; duplicate hosts are shown once. Missing and older-version measurements are excluded. How AA measures it
Snapshot 11 Nov 2026, 10:48 UTC · 355 matching endpoints.
Sources & methodologyOperational charts use every model with both selected measurements. Capability and task-cost charts require the same benchmark version. A model without an index can still have useful prices, speed or context data; it remains in the table and operational views.
One point per model uses an exact configuration and a host with both plotted measurements when available, then the lowest recorded input price and output-price tie-breaker. Select a model to see and compare its other hosts. Every provider endpoint is also available through the Show control.
Colors identify creators or providers. Labels name selected and leading points; they never change the data. The efficient frontier marks tradeoffs that no other plotted point improves in both dimensions. Log scales spread wide ranges; known zero prices remain visible. Sources & methodology.