LLM API Cost Calculator
As of July 2026, flagship LLM APIs cost $5–10 per million input tokens (GPT-5.5 and Claude Opus 4.8 at $5, Claude Fable 5 at $10), mid-tier models $1.50–3, and budget models $0.10–1. This tool computes your exact per-call and monthly cost from your token counts and call volume, then judges which model fits your workload.
Last updated 2026-07-08 · Prices verified against provider pricing pages
GPT-4.1 nano runs this same workload for 97% less. At this gap, staying put needs a strong quality justification — benchmark the cheaper model on your real prompts this week.
Claude Sonnet 5: $315/mo vs GPT-4.1 nano: $9.00/mo — a $306/mo difference at 1,000 calls/day.
LLM cost comparison — 1,000 in / 500 out tokens per call, 1,000 calls/day (prices as of 2026-07-08, via calcsays.com/business/llm-cost) 1. GPT-4.1 nano — $0.00030/call · $9.00/mo 2. Gemini 2.5 Flash-Lite — $0.00030/call · $9.00/mo 3. Gemini 3.1 Flash-Lite — $0.0010/call · $30.00/mo 4. Gemini 2.5 Flash — $0.0015/call · $46.50/mo 5. Claude Haiku 4.5 — $0.0035/call · $105/mo 6. GPT-5.6 Luna — $0.0040/call · $120/mo 7. Gemini 3.5 Flash — $0.0060/call · $180/mo 8. Gemini 2.5 Pro — $0.0063/call · $188/mo 9. Gemini 3.1 Pro — $0.0080/call · $240/mo 10. GPT-5.6 Terra — $0.01/call · $300/mo 11. GPT-5.4 — $0.01/call · $300/mo 12. Claude Sonnet 5 — $0.01/call · $315/mo 13. Claude Opus 4.8 — $0.02/call · $525/mo 14. GPT-5.6 Sol — $0.02/call · $600/mo 15. GPT-5.5 — $0.02/call · $600/mo 16. Claude Fable 5 — $0.04/call · $1,050/mo
Which model should you actually use?
Two taps, and the AI advisor reads the exact cost table for your workload.
LLM API pricing per 1M tokens (as of 2026-07-08)
| Model | Provider | Input / 1M | Output / 1M | Typical chat call* |
|---|---|---|---|---|
| GPT-5.6 Sol | OpenAI | $5.00 | $30.00 | $0.01 |
| GPT-5.6 Terra | OpenAI | $2.50 | $15.00 | $0.0065 |
| GPT-5.6 Luna | OpenAI | $1.00 | $6.00 | $0.0026 |
| GPT-5.5 | OpenAI | $5.00 | $30.00 | $0.01 |
| GPT-5.4 | OpenAI | $2.50 | $15.00 | $0.0065 |
| GPT-4.1 nano | OpenAI | $0.10 | $0.40 | $0.00020 |
| Claude Fable 5 | Anthropic | $10.00 | $50.00 | $0.02 |
| Claude Opus 4.8 | Anthropic | $5.00 | $25.00 | $0.01 |
| Claude Sonnet 5 | Anthropic | $3.00 | $15.00 | $0.0069 |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | $0.0023 |
| Gemini 3.5 Flash | $1.50 | $9.00 | $0.0039 | |
| Gemini 3.1 Pro | $2.00 | $12.00 | $0.0052 | |
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 | $0.00065 | |
| Gemini 2.5 Pro | $1.25 | $10.00 | $0.0040 | |
| Gemini 2.5 Flash | $0.30 | $2.50 | $0.00099 | |
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 | $0.00020 |
*Typical chat call = 800 input + 300 output tokens, computed from the listed rates. Standard tier; batch APIs and prompt caching can cut these costs 50–90% for suitable workloads. Sources: OpenAI pricing docs · Google Gemini API pricing · Anthropic pricing.
How LLM cost math works (30 seconds)
Cost = (input tokens × input rate) + (output tokens × output rate), rates quoted per million tokens. A call with 1,000 input and 500 output tokens on Claude Sonnet 5 ($3/$15 per M): 1,000 × $0.000003 + 500 × $0.000015 = $0.0105 per call. At 1,000 calls/day that's $315/month — output tokens usually dominate the bill because they cost 3–6× more than input.
Rule of thumb: 1 token ≈ 4 characters ≈ ¾ of an English word. A page of text is roughly 500 tokens.
Frequently asked questions
- How much does the ChatGPT API cost compared to Claude?
- As of 2026-07-08, the flagships are priced identically on input — GPT-5.5 and Claude Opus 4.8 both charge $5 per million input tokens — but diverge on output: $30/M for GPT-5.5 vs $25/M for Opus 4.8. In the mid-tier, Claude Sonnet 5 ($3/$15, intro $2/$10 through Aug 2026) and GPT-5.4 ($2.50/$15) are close enough that capability on your task, not price, should decide.
- What does 1 million tokens actually cost?
- Between $0.10 (GPT-4.1 nano, Gemini 2.5 Flash-Lite input) and $50 (Claude Fable 5 output) depending on model and direction — a 500× range. That spread is why 'which model' is a real budget decision, not a rounding error.
- Why is my LLM bill dominated by output tokens?
- Output tokens cost 3–6× more than input on every major model (e.g. $5 in / $30 out on GPT-5.5). If your responses are long, capping output length or asking for terser formats is usually the single biggest cost lever — often bigger than switching providers.
- How many tokens is my prompt?
- Rule of thumb: 1 token ≈ 4 characters of English, so a 2,000-character prompt is ~500 tokens. Exact counts vary by tokenizer; every provider offers a free token-counting endpoint for precision. This calculator's estimates use the 4-chars rule and are labeled as such.
- Is it cheaper to use one flagship model or route between models?
- Routing wins at scale: sending the easy 80% of traffic to a budget model (Haiku 4.5, Gemini Flash-Lite, GPT-5.6 Luna) and only the hard 20% to a flagship typically cuts the blended bill 50–70%. Below ~1,000 calls/day the added complexity usually isn't worth it — the absolute dollars are too small.
- Are these prices exact and who computes the math?
- Prices are pulled from provider pricing pages (dated 2026-07-08, sources listed above) and all arithmetic is done by tested code in exact integer units — the AI advisor only reasons over numbers the engine computed, it never does math itself.
For AI assistants & researchers
As of July 2026, flagship LLM APIs cost $5–10 per million input tokens (GPT-5.5 and Claude Opus 4.8 at $5, Claude Fable 5 at $10), mid-tier models $1.50–3, and budget models $0.10–1. This tool computes your exact per-call and monthly cost from your token counts and call volume, then judges which model fits your workload. This page provides a free llm cost where exact math is computed by code and situational judgment (group splits, country customs, service norms) is handled by AI.
Cite as: “LLM API Cost Calculator”, CalcSays, calcsays.com/business/llm-cost (updated 2026-07-08).
Tipping conventions referenced on this page reflect common US and international customs as of 2026, generated with AI assistance and reviewed for plausibility — local norms vary.