Price of a finished task
What a successful task actually costs
DeepSeek V4.1 Flash scores 81.1 at $0.029 a task. The leader, Claude Fable 5.1, scores 83.4 at $1.212 — 42× the price for 2.3 points.
Price per task
Cheapest first. Scores stay so the price has a quality next to it.
| # | Model | $/task | Overall | Reasoning | Coding | Agentic | Math | Data | Language | Instruct |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | DeepSeek V4.1 Flash Max open | 0.029 | 81.1 | 86.7 | 80.0 | 77.3 | 93.3 | 79.3 | 81.2 | 70.0 |
| 2 | Gemini 3.7 Flash High | 0.157 | 78.8 | 87.8 | 78.9 | 58.3 | 93.5 | 68.0 | 85.5 | 79.9 |
| 3 | Grok 4.6 | 0.207 | 78.0 | 90.5 | 76.8 | 57.0 | 92.6 | 73.9 | 83.7 | 71.9 |
| 4 | Muse Spark 1.3 xHigh | 0.219 | 81.6 | 89.7 | 81.1 | 64.1 | 95.9 | 79.6 | 82.8 | 78.0 |
| 5 | Qwen 3.8 Max open | 0.275 | 78.5 | 88.2 | 72.9 | 64.6 | 91.3 | 78.4 | 79.7 | 74.1 |
| 6 | Kimi K3 open | 0.348 | 79.2 | 90.7 | 81.4 | 62.2 | 84.4 | 78.7 | 85.5 | 71.4 |
| 7 | GPT-5.6 Terra Max | 0.352 | 77.9 | 90.6 | 78.2 | 54.9 | 94.9 | 79.3 | 82.9 | 64.6 |
| 8 | Muse Spark 1.2 xHigh | 0.375 | 78.0 | 90.0 | 77.5 | 57.6 | 91.2 | 76.5 | 78.6 | 74.3 |
| 9 | GPT-5.4 Thinking xHigh | 0.387 | 78.0 | 88.1 | 77.5 | 53.8 | 94.1 | 79.3 | 82.6 | 70.2 |
| 10 | GPT-5.5 Thinking xHigh | 0.435 | 80.2 | 89.7 | 82.1 | 54.0 | 95.9 | 81.6 | 87.4 | 70.7 |
| 11 | GPT-5.6 Sol Max | 0.515 | 81.0 | 91.7 | 83.9 | 56.2 | 96.2 | 79.8 | 87.7 | 71.8 |
| 12 | Claude 5 Opus Thinking Max | 0.699 | 80.1 | 91.2 | 81.4 | 65.2 | 95.7 | 74.6 | 88.7 | 63.8 |
| 13 | GPT-6 Astra Max | 0.736 | 82.2 | 92.7 | 80.4 | 57.3 | 96.8 | 83.0 | 89.4 | 75.6 |
| 14 | Claude Fable 5.1 Max | 1.212 | 83.4 | 91.7 | 86.4 | 66.1 | 97.0 | 80.3 | 89.5 | 73.0 |
| 15 | Claude Fable 5 Max | 1.439 | 83.0 | 89.7 | 86.0 | 62.2 | 96.0 | 80.5 | 90.7 | 75.8 |
Source: LiveBench release 2026-06-25. Cost per successful task = (Σ cost ÷ Σ questions ÷ score) × 100, USD. Green — three cheapest, orange — three most expensive. Category columns are scores, not prices.