Task cost calculator
The cost of one AI task, then per hour, week and month: input that grows every step, cache reads and writes, visible output and hidden reasoning. Anthropic: Claude Sonnet 5.5 costs $0.348 per L2 task (a 50,000-token document analysed in 3 steps) at list prices. With the prompt cache on (75 % of step 1 cached, earlier context read from cache on later steps) at its listed cache prices, the same task costs $0.082.
Prices as listed in the United States on . Runs in your browser. Your inputs stay on this device. All calculators · API prices · Price change briefings
Result Anthropic: Claude Sonnet 5.5 · L2 · 20 tasks per hour · data of 9 Nov 2026
- Per task
- $0.348
- Per hour
- $6.96
- Per week · 168 h
- $1,169.28
- Per month · 730 h
- $5,080.80
Office hours: $278.40 a week (40 h) · $1,206.38 a month (173.33 h).
- Cache: off (0 % cached). This model lists cache prices: reads $0.10, writes $2.50 (5-minute lifetime) per 1M tokens, from the provider’s price list read 9 Nov 2026.
- Long context: no higher long-context price for this model in our data.
- Context window: largest step 55,500 input + 1,000 output = 56,500 tokens of 1,000,000 — fits
- Rate limit: on average 53,000 input and 1,000 output tokens per minute at 20 tasks per hour (cached reads do not count, as Anthropic states): within the Start tier of Anthropic at an even pace (published limits read 17 Oct 2026); bursts within a minute are higher. Published tier limits
Standard tasks, cost per task
Visible output only; hidden reasoning is not included. L1, L2 and L3: list prices with no cache. The two cached columns use the share of each scenario (L2 75 %, L3 85 % of the step 1 input from cache, the whole earlier context from cache on later steps, 20 tasks an hour) at the provider’s listed cache prices, for models whose provider lists them. Prices as listed on 9 Nov 2026.
| Model | L1 | L2 | L3 | L2 · 75 % cached | L3 · 85 % cached | Price from |
|---|---|---|---|---|---|---|
| Anthropic: Claude Haiku 5.5 | $0.0003 | $0.017 | $0.424 | $0.0048 | $0.082 | provider price |
| Anthropic: Claude Opus 5.5 | $0.010 | $0.696 | $5.44 | $0.164 | $0.904 | provider price |
| Anthropic: Claude Sonnet 5.5 | $0.0050 | $0.348 | $2.72 | $0.082 | $0.452 | provider price |
| DeepSeek: DeepSeek V4 Pro 0813 | $0.0028 | $0.222 | $1.77 | $0.041 | $0.218 | provider price |
| DeepSeek: DeepSeek V4.1 Flash 1017 | $0.0007 | $0.051 | $0.405 | $0.0097 | $0.047 | provider price |
| Google: Gemini 3.8 Flash | $0.0019 | $0.131 | $1.02 | $0.035 | $0.198 | provider price |
| Google: Gemini 4 Argon | $0.0050 | $0.348 | $2.72 | cache price not in our data | cache price not in our data | provider price |
| Mistral: Mistral Large 4 | $0.0015 | $0.118 | $0.940 | $0.029 | $0.172 | provider price |
| MoonshotAI: Kimi K3 | $0.0075 | $0.522 | $4.08 | $0.140 | $0.792 | provider price |
| OpenAI: GPT-5.4 Mini | $0.0020 | $0.133 | $1.03 | cache price not in our data | cache price not in our data | provider price |
| OpenAI: GPT-6 Astra | $0.025 | $1.74 | $13.61 | $0.480 | $2.87 | provider price |
| OpenAI: GPT-6 Sol | $0.0050 | $0.348 | $2.72 | $0.096 | $0.574 | provider price |
| Qwen: Qwen3.8 Max (0902) | $0.0042 | $0.336 | $2.69 | cache price not in our data | cache price not in our data | provider price |
| xAI: Grok 4.5 | $0.0042 | $0.336 | $2.69 | $0.096 | $0.614 | OpenRouter |
| Z.ai: GLM 5.3 | $0.0030 | $0.236 | $1.88 | $0.075 | $0.493 | provider price |
Assumptions
| What | Value | Why |
|---|---|---|
| Tokenizer factor | 1.00 | Token counts are taken as entered; provider tokenizers differ. |
| Retries | 1.00 | No repeated requests for broken output. |
| Cache | your share, 0 % by default | Cache read and write prices come from the provider’s price page where it lists them, with the date it was read; a price it does not list is charged at the input price. Step 1 reads your cached share when the cache is still warm from the previous task and writes it otherwise. From step 2 the whole earlier context of the task is read from cache and only the new tokens of the step are written, as providers bill a loop that resends its history; where the provider lists no write price, the new tokens cost the input price. At 0 % nothing is cached. Cache storage fees (Gemini) are not included. |
| Cache lifetime | 5 minutes or 1 hour | Your setting. The cache is warm for the next task when tasks come more often than this. Anthropic lists separate write prices for 5 minutes and 1 hour; other price pages in our data do not state a lifetime. |
| Long context | provider’s price | Where the provider charges more above a token threshold, every step above it is priced at the higher rate: input, cache and output. |
| Hidden reasoning | your estimate | Enter tokens per step; they are charged at the output price. |
| Month | 730 h | Round-the-clock workload; office hours would be 173 h. |
Formula: for each step, input = input at step 1 + growth × (step − 1) + tool schema; cost = fresh input × input price + cache writes × cache-write price + cache reads × cache-read price + (visible output + hidden reasoning) × output price, per million tokens. The methodology explains how prices are read; API prices: caching and batch compares cache and long-context prices by model.