Price against quality
The frontier
Every scored model, plotted against what it costs to run. The models on the step are the ones nothing else beats on both price and score. Read it left to right: the cheapest model that clears the quality you need is the one you want.
Overall percentile against output price
84 models plotted · 9 on the frontier · firm evidence only
- On the frontier
- Beaten on both
- Open weights
Output price, $ per million tokens · log scale →↑ Overall percentile
Cheapest on the frontier
Best score at any price
On the frontier
| Model | Lab | Composite | Out $/M | In $/M | Evidence |
|---|---|---|---|---|---|
| GPT-6 Astra | OpenAI | 99th | $50.00 | $10.00 | 15 bench · 2 src |
| GPT-6.1 Sol | OpenAI | 99th | $10.00 | $2.00 | 13 bench · 2 src |
| Gemini 3.7 Flash | 89th | $7.50 | $1.50 | 9 bench · 2 src | |
| Qwen3.8 Max | Alibaba | 88th | $6.00 | $2.00 | 8 bench · 2 src |
| DeepSeek V4 Pro 0813 | DeepSeek | 86th | $3.96 | $1.32 | 8 bench · 2 src |
| DeepSeek V4.1 Flash | DeepSeek | 84th | $1.20 | $0.30 | 9 bench · 2 src |
| Kimi K3 | Kimi | 78th | $1.00 | $0.30 | 9 bench · 2 src |
| Claude Haiku 5.5 | Anthropic | 75th | $0.50 | $0.10 | 10 bench · 2 src |
| DeepSeek V4 Flash 0731 | DeepSeek | 70th | $0.28 | $0.14 | 8 bench · 2 src |
Price is the catalog list price for output tokens, from the model's own vendor where it has a listing. A model with no price in the catalog cannot be plotted. See the full ranking.