11-Model API Cost Comparison (2,000 Input + 800 Output Tokens)
| Model | Cost / 1K Calls | Monthly Bill (120,000 Calls) |
|---|---|---|
| Anthropic Claude 4 Sonnet / 3.7 Sonnet | $18.00 | $2,160.00 / mo |
| OpenAI GPT-5 / GPT-4.1 Flagship | $10.40 | $1,248.00 / mo |
| OpenAI GPT-4o Multimodal | $13.00 | $1,560.00 / mo |
| OpenAI GPT-4o mini (High-Speed) | $0.78 | $93.60 / mo |
| OpenAI o3-mini / o4-mini (Reasoning) | $5.72 | $686.40 / mo |
| Anthropic Claude 4 Opus | $90.00 | $10,800.00 / mo |
| Anthropic Claude 3.5 Haiku | $4.80 | $576.00 / mo |
| Google Gemini 2.5 Pro (2M Context) | $10.50 | $1,260.00 / mo |
| Google Gemini 2.5 Flash (1M Context) | $0.78 | $93.60 / mo |
| DeepSeek R1 / V3 Reasoning API | $1.42 | $170.40 / mo |
| Meta Llama 4 70B / 405B (Groq / Together) | $1.81 | $217.44 / mo |
Frequently Asked Questions
How does OpenAI Prompt Caching reduce GPT-5 and GPT-4o input token costs? ▾
OpenAI automatically caches repeated prompt prefixes longer than 1,024 tokens (such as system instructions, few-shot examples, and tool definitions), discounting cached input tokens by 50% to 75%.
Are reasoning tokens in o3-mini and o4-mini billed as input or output tokens? ▾
OpenAI bills internal chain-of-thought reasoning tokens at the model’s output token rate ($4.40 per 1M tokens on o3-mini), even though they are hidden from the final API text response.