LLM API Fee Calculator (2026 Updated Rates)
LLM API costs look small per call — until you multiply 2,000 input + 500 output tokens by thousands of calls a day. This calculator prices OpenAI, Anthropic, Google and DeepSeek models with prompt-caching hit rates and the 50% Batch API discount, then projects daily, monthly and yearly spend.
LLM API fee quick reference (2026)
| Model | List price |
|---|---|
| OpenAI GPT-5 | $1.25 in / $10 out per 1M |
| OpenAI GPT-5 mini | $0.25 in / $2 out per 1M |
| OpenAI GPT-4.1 | $2 in / $8 out per 1M |
| OpenAI GPT-4o | $2.5 in / $10 out per 1M |
| OpenAI o3 | $2 in / $8 out per 1M |
| OpenAI o4-mini | $1.1 in / $4.4 out per 1M |
| Anthropic Claude Opus 4.5 | $5 in / $25 out per 1M |
| Anthropic Claude Sonnet 4.5 | $3 in / $15 out per 1M |
| Anthropic Claude Haiku 4.5 | $1 in / $5 out per 1M |
| Anthropic Claude Opus 4.1 | $15 in / $75 out per 1M |
| Google Gemini 2.5 Pro | $1.25 in / $10 out per 1M |
| Google Gemini 2.5 Flash | $0.3 in / $2.5 out per 1M |
| DeepSeek DeepSeek V3.2 | $0.28 in / $0.42 out per 1M |
| Monthly budget | Max calls/day* |
|---|---|
| $50.00 | 123 |
| $100.00 | 246 |
| $250.00 | 617 |
| $500.00 | 1,234 |
| $1,000.00 | 2,469 |
* 2,000 input + 500 output tokens per call, no cache.
LLM API fee formula & hidden costs
Per-token pricing across OpenAI, Anthropic, Gemini & DeepSeek
Cost = calls × (input×(1−cache)×P_in + input×cache×P_cache + output×P_out)/1M × (batch ? 0.5 : 1).
Hidden fees to watch
- Cached input tokens bill at ~10% of list — a 50% cache hit rate cuts input cost almost in half.
- Batch API halves the bill for non-urgent workloads (24h SLA).
- Reasoning models bill their hidden "thinking" tokens as output tokens.
- Retries and tool-call loops multiply token spend faster than any rate card suggests.
How to pay less in LLM API fees
- Use the reverse calculator to pass fees to the client explicitly instead of eating them.
- Batch small orders — fixed per-order fees dominate sub-$10 transactions.
- Avoid currency conversion inside the platform; settle in your own currency via a multi-currency account.
- Compare Claude Skills Club — Community and tooling for Claude power users.
Frequently asked questions
How much does 1,000 calls/day cost on Claude Sonnet 4.5?
At 2,000 input + 500 output tokens per call ($3/$15 per 1M): $0.0135/call = $13.50/day, ~$405/month. A 50% cache hit rate drops it to ~$324/mo.
How much does prompt caching actually save?
Cached input tokens bill at roughly 10% of list price. At a 50% hit rate on Sonnet 4.5, the input leg costs $3x0.5 + $0.30x0.5 = $1.65/1M instead of $3.
What is the cheapest frontier model here?
DeepSeek V3.2 at $0.28/$0.42 per 1M undercuts even Gemini 2.5 Flash ($0.30/$2.50) — often 10-50x cheaper than flagship models.
When is the Batch API worth it?
Any non-interactive workload — evals, embeddings, nightly summaries — saves 50% by tolerating up to 24h latency.
How many calls/day can I run on a $300/month budget?
Use reverse mode: at Sonnet 4.5 with 2k in / 500 out and no cache, $300 buys roughly 740 calls/day; with a 50% cache hit it buys ~926.