Capability per dollar
73 benchmarked models plotted by capability against blended price. The 11 models on the frontier are the ones no other model beats on both numbers at once. There is no combined score: capability and price stay separate, always.
Frontier (11)All benchmarked models (73)
The frontier, cheapest first
| Model | Capability (ECI) | Blended price |
|---|---|---|
 Llama 3.1 8BMeta | 50.3 | $0.06/M |
 Phi 4Microsoft | 64.6 | $0.09/M |
 GPT 5 NanoOpenAI | 73.4 | $0.14/M |
 DeepSeek V4 ProDeepSeek | 81.6 | $0.26/M |
 GLM 5.2Zhipu AI | 83.4 | $1.07/M |
 Grok 4.20xAI | 85.7 | $1.56/M |
| Grok 4.5xAI | 88.5 | $3.00/M |
 Claude Opus 4.8Anthropic | 89.2 | $10.00/M |
| GPT 5.5OpenAI | 89.6 | $11.25/M |
| Claude Fable 5Anthropic | 91.6 | $20.00/M |
| GPT 5.5 ProOpenAI | 91.8 | $67.50/M |
How this is computed
Capability is the Epoch Capabilities Index at each model's peak reasoning tier (0 to 100, CC-BY data). Price is the blended list price per 1M tokens at a 3:1 input:output mix, from provider list pricing. Models missing either number are not plotted rather than shown as zero. A model is on the frontier when no other model is at least as capable and at least as cheap.
Related boards: cheapest models, best free models and measured speed.