LLM API Pricing Comparison
List prices per million tokens, pulled from official pricing pages. Output tokens cost 4–8× more than input on most models — that, not the headline rate, is what drives your bill.
| Model | Input $/1M | Cache hit $/1M | Output $/1M | Output÷Input |
|---|---|---|---|---|
| OpenAI · GPT-6 Astra | $10.000 | $1.0000 | $50.00 | 5.0× |
| OpenAI · GPT-6.1 Sol | $2.000 | $0.1000 | $10.00 | 5.0× |
| OpenAI · GPT-6.1 Terra | $1.000 | $0.0500 | $5.00 | 5.0× |
| OpenAI · GPT-6 Luna | $0.100 | $0.0100 | $0.50 | 5.0× |
| Anthropic Claude · Claude Haiku 5.5 | $0.100 | — | $0.50 | 5.0× |
| Anthropic Claude · Claude Sonnet 5 | $2.000 | $0.2000 | $10.00 | 5.0× |
| Anthropic Claude · Claude Opus 5 | $5.000 | $0.5000 | $25.00 | 5.0× |
| Google Gemini · Gemini 3.8 Flash | $0.750 | $0.0750 | $3.75 | 5.0× |
| Google Gemini · Gemini 3.1 Pro | $2.000 | $0.2000 | $12.00 | 6.0× |
| Google Gemini · Gemini 3.5 Flash-Lite | $0.300 | $0.0300 | $2.50 | 8.3× |
| DeepSeek · DeepSeek V4.1 Flash | $0.140 | $0.0028 | $0.28 | 2.0× |
| DeepSeek · DeepSeek V4 Pro | $0.435 | $0.0036 | $0.87 | 2.0× |
Verified 2026-11-07. Cache-hit prices apply to repeated prefixes (system prompts, tool definitions). DeepSeek prices shown are off-peak; peak hours double.
What actually drives your bill
Three levers move cost far more than choosing between adjacent models: output length (output is priced 4–8× input), cache hit rate (prompt caching cuts input cost up to 98% on some providers), and request volume. Before optimizing your model choice, measure these three from your usage dashboard.