MundoKit

LLM API cost calculator

Enter tokens per request and your request volume to estimate a token budget. Choose a dated model preset or enter custom prices; blank overrides use the selected preset.

Inputs
Includes cached reads and cache writes; do not add them twice.
Use actual cache hits, not just tokens you hope to cache.
A separate input category. OpenAI presets use 30-minute writes; Claude presets use 5-minute writes.
Use the total billed output count, including thinking; do not add reasoning again if already included.
An assumed workload, not a provider limit.
Editable planning assumption.
Percent applied to all token charges and comparison rows. This is a scenario assumption, not a claim of batch support or cache discount stacking for every provider.
USD / 1M
Blank uses the preset’s applicable context tier. Custom requires all four rates. Enter 0 for a free rate.
USD / 1M
Blank uses the preset’s applicable context tier. Custom requires all four rates. Enter 0 for a free rate.
USD / 1M
Blank uses the preset’s applicable context tier. Custom requires all four rates. Enter 0 for a free rate.
USD / 1M
Blank uses the preset’s applicable context tier. Custom requires all four rates. Enter 0 for a free rate.

Results

Estimated monthly token cost
$23.40
Per request
$0.00078
Per day
$0.78
Applied rates: input / read / write / output
$0.1 / $0.01 / $0.125 / $0.5 per 1M
Preset comparison, 2026-11-14; same workload and assumed 0% discount. Overrides affect only the result above.
ModelTierInput / read / write / output ($/1M)Per requestPer month
Anthropic Claude Haiku 5.5Base0.1 / 0.01 / 0.125 / 0.5$0.00078$23.40
OpenAI GPT-6 LunaBase0.1 / 0.01 / 0.125 / 0.5$0.00078$23.40
DeepSeek V4.1 Flash (off-peak)Base0.15 / 0.003 / 0.15 / 0.6$0.000924$27.72
Mistral Small 4Base0.15 / 0.015 / 0.15 / 0.6$0.00102$30.60
DeepSeek V4.1 Flash (peak)Base0.3 / 0.006 / 0.3 / 1.2$0.00185$55.44
Google Gemini 3.5 Flash-Lite (implicit cache)Base0.3 / 0.03 / 0.3 / 2.5$0.00334$100.20
DeepSeek V4 Pro (off-peak)Base0.66 / 0.022 / 0.66 / 1.98$0.00348$104.28
Mistral Large 4Base0.7 / 0.07 / 0.7 / 2.1$0.00406$121.80
DeepSeek V4 Pro (peak)Base1.32 / 0.044 / 1.32 / 3.96$0.00695$208.56
Anthropic Claude Sonnet 5.5Base2 / 0.1 / 2.5 / 10$0.01$444.00
Google Gemini 4 Carbon (November promotion)Base2 / 0.1 / 2 / 10$0.01$444.00
OpenAI GPT-6.1 SolBase2 / 0.1 / 2.5 / 10$0.01$444.00
Google Gemini 4 Carbon (from December 1, 2026)Base4 / 0.2 / 4 / 20$0.03$888.00
Anthropic Claude Fable 5.2Base8 / 0.2 / 10 / 40$0.06$1,728.00
OpenAI GPT-6 AstraBase10 / 1 / 12.5 / 50$0.08$2,340.00
  • Base preset tier selected using total input, including reads and writes. Rate overrides replace the final tier rates.
  • Token costs only: excludes taxes, tool calls, explicit cache storage, regional premiums and subscription charges. This comparison does not validate model context limits or equivalent output quality.
  • Choose the Carbon preset for your budget dates: November promotion or rates from December 1, 2026. DeepSeek has separate peak and off-peak rows. All prices are editable.

Results update as you type. The share link stores your inputs after the # in the address, which is not sent to our server, but anyone you send it to can read them.

Prices change often. We list the date we last checked each price; confirm on the provider’s official pricing page before you budget.

How it works

Provider pricing tables quote USD per million tokens. Split total input I into cached reads C, cache writes W and ordinary input I − C − W. Each token belongs to just one category. The OpenAI caching guide confirms that the cache-write price is a total rate, not an extra fee on top of ordinary input.

request cost = [(I − C − W) × input rate + C × read rate + W × write rate + output × output rate] ÷ 1,000,000 × (1 − discount ÷ 100) monthly cost = request cost × monthly requests daily cost = monthly cost ÷ chosen days

For daily request volume, monthly requests equal daily requests times the chosen days. A 30-day month is an editable budget assumption. The batch percentage is a user-supplied scenario discount applied uniformly, including in the comparison. Verify the provider’s batch and caching combination before entering a discount. Preset rows keep their published prices even when you override the selected result.

OpenAI Astra, Sol and Luna switch the whole request to higher prices above 272,000 input tokens. Claude Sonnet 5.5 has no long-prompt surcharge. Claude Haiku 5.5 switches above 100,000 input tokens. Cached reads and writes count toward those thresholds. Overrides replace the resulting tier price. Google Carbon has separate presets for its November promotion and the published rates from December 1, 2026: $4 input, $0.20 cached input and $20 output per million tokens. Both appear in the comparison; choose the one matching your budget dates. DeepSeek has peak and off-peak presets. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday–Friday excluding Chinese public holidays. All other hours, including weekends and Chinese public holidays, use half-price off-peak rates. No intermediate rounding is performed. Displayed money uses cents, or three significant digits for nonzero amounts below one cent.

Worked examples

A daily workload with cached reads

GPT-6 Luna with 10,000 input tokens, including 8,000 cached reads and no writes, plus 1,000 output tokens costs (2,000 × $0.10 + 8,000 × $0.01 + 1,000 × $0.50) ÷ 1,000,000 = $0.00078 per request. At 1,000 requests per day for 30 days, that is $0.78 per day and $23.40 per month.

Cache writes and an assumed batch discount

GPT-6 Astra with 10,000 total input, 8,000 reads, 1,000 writes and 1,000 output tokens costs (1,000 × $10 + 8,000 × $1 + 1,000 × $12.50 + 1,000 × $50) ÷ 1,000,000 = $0.0805. At 1,000 daily requests for 30 days it costs $2,415. A 50% batch scenario gives $1,207.50 per month.

Crossing a long-prompt threshold

Claude Haiku 5.5 with 100,001 ordinary input tokens, no reads or writes, and 1,000 output tokens uses $0.50 input and $2.50 output rates. The unrounded request cost is $0.0525005, displayed as $0.05. Exactly 100,000 input tokens would cost $0.0105 before display rounding.

Frequently asked questions

Are cached tokens additional to input tokens?
No. Enter the total prompt count, then identify the portions billed as cached reads and writes. Their sum cannot exceed total input. Ordinary input is the remainder.
Do reasoning tokens cost money?
Yes. OpenAI bills reasoning tokens as output, and Gemini’s output prices include thinking tokens. Use the billed total output count. If reasoning is already included in that count, do not add it a second time.
Can I change a preset’s price?
Yes. Enter final USD-per-million rates in the override fields. Zero is accepted; blank means use the preset. Custom mode requires all four rates. Overrides affect your result, while the comparison retains its dated preset prices.
Does this estimate my complete API bill?
It estimates text token charges. It excludes tools, grounding, explicit cache storage, taxes, regional premiums and subscriptions. Gemini Flash-Lite comparisons assume implicit caching. Actual token counts and cache hits can differ by model.
Does the comparison guarantee batch discounts or model capacity?
No. The discount is a hypothetical uniform percentage, and the table compares prices at identical token counts. It does not validate context windows, request limits, output quality or whether a provider supports your chosen batch and caching combination.

Sources

Last reviewed 2026-11-14; next review due by 2026-11-30. The method is written out above so you can check our math.