Best-value AI models
The frontier contains models for which no eligible alternative is at least as capable and no more expensive, with a strict advantage in one of those measures. It helps you compare trade-offs between benchmark quality and cost.
Which should you choose?
No single benchmark covers every use case. Try your own prompts and compare accuracy, response time, input/output support, and the actual bill.
AI models
| Model | Intelligence score | Benchmark cost (USD) | Input price / 1M tokens (USD) | Output price / 1M tokens (USD) | Output speed |
|---|---|---|---|---|---|
| Granite 4.2 3BIBMBest value frontier | 9.10 | $14.9 | 0.03 | 0.12 | 216 tok/s |
| MiMo-V2.5XiaomiBest value frontier | 22.30 | $42.6 | 0.14 | 0.28 | 65 tok/s |
| MiMo-V2.5-ProXiaomiBest value frontier | 26.40 | $138 | 0.43 | 0.87 | 36 tok/s |
| GPT-5.6 Luna (xhigh)OpenAIBest value frontier | 34.80 | $179 | 0.20 | 1.20 | 110 tok/s |
| GLM-5.3-FlashZ AIBest value frontier | 41.90 | $280 | 0.15 | 0.50 | 66 tok/s |
| GPT-5.6 Sol (high)OpenAIBest value frontier | 42.50 | $1487 | 4.00 | 20.00 | 62 tok/s |
| GPT-6 Astra (low)OpenAIBest value frontier | 46.00 | $1537 | 10.00 | 50.00 | 47 tok/s |
| Muse Spark 1.3 (max)MetaBest value frontier | 48.20 | $2000 | 1.25 | 4.25 | 229 tok/s |
| GPT-6 Astra (medium)OpenAIBest value frontier | 49.70 | $2434 | 10.00 | 50.00 | 47 tok/s |
| GPT-6 Astra (high)OpenAIBest value frontier | 51.00 | $2917 | 10.00 | 50.00 | 50 tok/s |
| GPT-6 Astra (xhigh)OpenAIBest value frontier | 52.50 | $3803 | 10.00 | 50.00 | 52 tok/s |
| GPT-6 Astra (max)OpenAIBest value frontier | 52.80 | $5324 | 10.00 | 50.00 | 56 tok/s |
| Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)AnthropicBest value frontier | 53.20 | $9063 | 10.00 | 50.00 | 60 tok/s |
| Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)AnthropicBest value frontier | 53.40 | $13129 | 10.00 | 50.00 | 69 tok/s |
Guides
How we compare models
Intelligence uses the Artificial Analysis Intelligence Index. Benchmark cost is the total USD cost of running that test, including reasoning tokens; it is not a per-token API price. Rankings use the same dataset and calculations in every language. General benchmark scores do not measure performance in a particular language or task.