LLM API Pricing Calculator 2026
Estimate your monthly and annual large language model API budgets in 2026. Instantly compare token costs across GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro, and DeepSeek V3 with prompt caching and Batch discounts.
Workload Definition
Configure model selection and monthly token volumes
User query, preamble, and cached document context sent to the API
Text tokens generated by the model (typically more expensive)
Batch API schedules non-urgent queries to run at flat half-price
Percentage of input tokens read from reused system templates or history cache
LLM Budget Summary
Annual & Monthly API expenditure
Enter your token volumes and select an LLM backend model, then click Estimate AI Token Budget to inspect your tailored cost profile.
Provider Pricing Grid (2026)
Standard list rates per million tokens (CPM) across primary developer API endpoints.
| AI Model & Endpoint | Input CPM (Base) | Cached Input CPM | Output CPM (Base) |
|---|---|---|---|
| OpenAI GPT-4o | $2.50 | $1.25 (50% Off) | $10.00 |
| OpenAI GPT-4o-mini | $0.15 | $0.075 (50% Off) | $0.60 |
| OpenAI o1 Reasoning | $15.00 | $7.50 (50% Off) | $60.00 |
| Claude 3.5 Sonnet | $3.00 | $0.30 (90% Off) | $15.00 |
| Claude 3.5 Haiku | $0.80 | $0.08 (90% Off) | $4.00 |
| Google Gemini 1.5 Pro | $1.25 | $0.625 (50% Off) | $5.00 |
| Google Gemini 1.5 Flash | $0.075 | $0.0375 (50% Off) | $0.30 |
| DeepSeek-V3 | $0.14 | $0.014 (90% Off) | $0.28 |
| DeepSeek-R1 | $0.55 | $0.055 (90% Off) | $2.19 |
| LLaMA 3.1 405B (Groq/API) | $2.66 | $2.66 (No discount) | $2.66 |
LLM Pricing FAQ
Learn more about optimizing token efficiency, prompt caching architectures, and selecting API options.
Prompt caching allows the LLM API to store frequently used context (like long system preambles, knowledge base files, or multi-turn chats) in fast-access memory. When a matching prompt is sent, the provider bills the cached tokens at a 50% to 90% discount. This is highly effective for agent loops and retrieval-augmented generation (RAG) tasks.
The Batch API allows you to submit lists of text completions to OpenAI, Anthropic, or Google Gemini as non-real-time jobs. The results are generated during low-traffic periods and returned within 24 hours. Because it fills off-peak capacity, providers offer a flat 50% discount on standard rates.
Input tokens include all instructions, user messages, and contextual files sent to the model. Output tokens are the response text generated by the model. Output tokens require significantly more computing overhead to produce (auto-regressive decoding), making them 3x to 5x more expensive than standard inputs.
DeepSeek’s pricing (V3 and R1) is highly disruptive. Standard DeepSeek-V3 input tokens cost $0.14 per million, reducing to $0.014 with caching hits, and output tokens are $0.28 per million. This is up to 90% cheaper than OpenAI’s standard GPT-4o ($2.50 input / $10.00 output CPM), making it highly attractive for extreme-scale text processing pipelines.
To minimize LLM API fees, structure your applications around: 1) Active prompt caching by keeping system instructions and custom databases identical; 2) Route non-urgent summarization, indexing, and offline logs tasks to Batch API channels; 3) Route classification and routing tasks to lighter models like GPT-4o-mini or Gemini 1.5 Flash, only reserving flagship models for reasoning-heavy prompts.