// kalkulator
What AI really costs
Pricing is given in dollars per million tokens, a unit nobody can feel. I convert them daily into three things you actually do: writing an article, a hundred chat queries, and an hour working with an agent that reads code. In USD.
rates pulled: 2026-11-12 06:10:17
| Model | Input / million | Output / million | 1,500-word article | 100 queries | Hour with an agent |
|---|---|---|---|---|---|
| Qwen 3.7 Flash | 0,03 $ | 0,13 $ | $0.0003 | $0.0073 | $0.01 |
| DeepSeek V4 Flash | 0,04 $ | 0,30 $ | $0.0007 | $0.02 | $0.02 |
| GPT-5.6 Luna Pro | 0,20 $ | 1,20 $ | $0.0029 | $0.06 | $0.09 |
| GPT-5.6 Luna | 0,20 $ | 1,20 $ | $0.0029 | $0.06 | $0.09 |
| Gemini 3.5 Flash Lite | 0,30 $ | 2,50 $ | $0.0060 | $0.12 | $0.15 |
| DeepSeek V4 Pro | 0,96 $ | 1,91 $ | $0.0056 | $0.14 | $0.30 |
| Gemini 3.7 Flash | 0,75 $ | 3,75 $ | $0.0094 | $0.20 | $0.30 |
| Gemini 3.6 Flash | 0,75 $ | 3,75 $ | $0.0094 | $0.20 | $0.30 |
| Qwen 3.7 Max | 1,48 $ | 4,43 $ | $0.01 | $0.28 | $0.50 |
| Kimi K3 | 0,19 $ | 16,00 $ | $0.04 | $0.65 | $0.53 |
| Gemini 3.5 Flash | 1,50 $ | 9,00 $ | $0.02 | $0.46 | $0.65 |
| Qwen 3.8 2.4T | 2,00 $ | 6,00 $ | $0.02 | $0.38 | $0.68 |
| Grok 4.6 | 2,00 $ | 6,00 $ | $0.02 | $0.38 | $0.68 |
| Grok 4.5 | 2,00 $ | 6,00 $ | $0.02 | $0.38 | $0.68 |
| Claude Sonnet 5 | 2,00 $ | 10,00 $ | $0.03 | $0.54 | $0.80 |
| GPT-5.6 Terra Pro | 2,00 $ | 12,00 $ | $0.03 | $0.62 | $0.86 |
| GPT-5.6 Terra | 2,00 $ | 12,00 $ | $0.03 | $0.62 | $0.86 |
| GPT-5.6 Sol Pro | 4,00 $ | 20,00 $ | $0.05 | $1.08 | $1.60 |
| GPT-5.6 Sol | 4,40 $ | 22,00 $ | $0.06 | $1.19 | $1.76 |
| Claude Opus 5 | 5,00 $ | 25,00 $ | $0.06 | $1.35 | $2.00 |
| GPT-6 Astra | 10,00 $ | 50,00 $ | $0.13 | $2.70 | $4.00 |
| Claude Fable 5.1 | 10,00 $ | 50,00 $ | $0.13 | $2.70 | $4.00 |
| Claude Fable 5 | 10,00 $ | 50,00 $ | $0.13 | $2.70 | $4.00 |
What I'm calculating
- 1,500-word article - 1 500 input tokens and 2 200 output, brief plus text in English
- 100 chat queries - 70 000 input tokens and 40 000 output, 700 tokens per question and 400 per answer
- Hour of work with an agent - 250 000 input tokens and 30 000 output, agent reads repository and writes fixes
- Rates apply tostandard API calls. Prompt caching, fast modes, and batch packages change the math in both directions.
- Reasoning models consume more output tokens than you see in the response. The table shows the rate, not the model's appetite.
- Where the half-rate for Fable 5 at Opus 5 came from, I explain in the Opus 5 hub.
// follow
New tests and price changes, straight to your feed reader
Free RSS, no account and no email needed.
Follow Quantslant →Sources: OpenRouter - provider rates. Table refreshes automatically every day.