Methodology and scope
In August 2026 we compared twelve frontier AI models using an intelligence index (AA) and a blended price per million tokens at a 3:1 input-to-output ratio. The data come from current public benchmarks and vendor price lists. The purpose is to test whether higher prices correspond to higher measured intelligence or whether some models offer a better price-to-performance ratio.
Intelligence leaders and price extremes
Claude Opus 5 from Anthropic leads the intelligence index at 60.7, followed closely by Claude Fable 5 at 59.9. Third place goes to GPT-5.6 Sol from OpenAI at 58.9. At the other end DeepSeek V4 Flash 0731 scores 49.9 but costs only 0.17 USD per million tokens.
On price, Claude Opus 5 is the most expensive at 10 USD per million tokens, followed by Claude Fable 5 at 20 USD. DeepSeek V4 Flash 0731 is the cheapest at 0.17 USD, an extreme gap versus the priciest models.
Price does not always track intelligence
The data show that the most expensive models are not always the most intelligent. For example GPT-5.6 Sol reaches 58.9 on the intelligence index for 11.25 USD, while Claude Fable 5 scores 59.9 for 20 USD. At the low end DeepSeek V4 Flash 0731 delivers 49.9 for just 0.17 USD, a vastly better cost efficiency.
Laboratory benchmarks can be misleading. The intelligence index measures theoretical performance; real-world deployment is also shaped by latency, stability and task-specific capabilities. Price does not always reflect practical utility for a given workload.
Implications for procurement
Organisations should assess whether a higher model price brings proportional added value. Models such as Claude Opus 5 or GPT-5.6 Sol provide top-tier performance but their cost may be disproportionate for many use cases. Alternatives such as DeepSeek V4 Flash 0731 or MiniMax-M3 deliver respectable performance at a fraction of the price.
The practical conclusion is that you do not always get what you pay for. Model selection should weigh intelligence index, cost efficiency and the specific requirements of the project. Outliers in the price-to-performance ratio can be compelling options for budget-sensitive applications.
Frequently asked questions
Which AI model has the highest intelligence index?
The highest intelligence index is held by Anthropic's Claude Opus 5 at 60.7. It is followed by Claude Fable 5 at 59.9 and OpenAI's GPT-5.6 Sol at 58.9. These models represent the pinnacle of measured intelligence, though their prices differ significantly.
Which model is the cheapest and what is its performance?
The cheapest model in the comparison is DeepSeek V4 Flash 0731 at a price of 0.17 USD per million tokens. Its intelligence index is 49.9, placing it among medium-performance models. The price/performance ratio is thus exceptional.
What is the price difference between the most expensive and cheapest model?
The assumed ratio of the most expensive model (Claude Fable 5 at 20 USD) to the cheapest model (DeepSeek V4 Flash 0731 at 0.17 USD) is 114.3×. This shows an extreme price spread between models with different performance levels.
DATA SOURCES AND METHOD
Všechna čísla v tomto srovnání pocházejí z uvedených zdrojů k datu 2 August 2026. Grafy generuje institut CIAD přímo ze zdrojových dat; textová analýza čísla nikdy nedopočítává ani neodhaduje.