ChatGPT API Cost Calculator 2026
Calculate exact OpenAI, Anthropic, and DeepSeek API token fees based on real-time 2026 pricing, prompt caching, and custom request volume.
💡 Related Calculators: Prompt Caching Savings Optimizer • Tokens to Words Converter • RAG Cost Calculator
2026 LLM API Pricing Benchmarks (Per 1M Tokens)
Official developer API pricing rates across OpenAI, Anthropic, and DeepSeek endpoints.
| Model | Standard Input | Cached Input | Output Generation | Context Window |
|---|---|---|---|---|
| DeepSeek-V3 | $0.14 | $0.014 (-90%) | $0.28 | 64k |
| DeepSeek-R1 (Reasoning) | $0.55 | $0.14 (-75%) | $2.19 | 64k |
| GPT-4o mini | $0.15 | $0.075 (-50%) | $0.60 | 128k |
| Claude 3.5 Haiku | $0.80 | $0.08 (-90%) | $4.00 | 200k |
| GPT-4o (Omni) | $2.50 | $1.25 (-50%) | $10.00 | 128k |
| Claude 3.5 Sonnet | $3.00 | $0.30 (-90%) | $15.00 | 200k |
| OpenAI o1 (Full Reasoning) | $15.00 | $7.50 (-50%) | $60.00 | 200k |
Frequently Asked Questions
How does prompt caching reduce API invoices by up to 90%?
When an application sends static system prompts, JSON schemas, or RAG reference documents larger than 1,024 tokens, modern providers cache the attention states in memory. Subsequent calls matching that prompt prefix skip raw compute, discounting input costs by 50% (OpenAI) or 75% to 90% (Anthropic and DeepSeek).
What makes output token generation 3x to 4x more expensive than prompt ingestion?
Prompt processing runs in parallel across GPU matrix cores in a single forward pass. Generation is strictly autoregressive, requiring one sequential GPU memory transfer cycle per generated token. This memory bandwidth bottleneck makes output generation dramatically costlier to host.
How are hidden reasoning tokens billed in models like DeepSeek-R1 and OpenAI o1?
Reasoning models generate internal Chain-of-Thought (CoT) tokens before producing visible output. Although these tokens are invisible in the API response or chat UI, providers bill them at the full output generation token rate, which can double or triple total request cost on complex logical queries.