What an AI API really costs in Latin American currencies (September 2026)

1. One million tokens in six currencies#

AI APIs are billed in dollars per million tokens. For buyers in Colombia, Mexico, Brazil, Chile, Argentina or Peru, the real cost depends on the exchange rate. The table uses Codifly's blended price (3 input tokens per 1 output token, no caching or batch) from the open dataset reviewed on 2026-09-08 against each provider's official page, converted at the reference rate of 2026-09-29: 1 USD = 3,301 COP; 1 USD = 17.90 MXN; 1 USD = 5.21 BRL; 1 USD = 962 CLP; 1 USD = 1,529 ARS; 1 USD = 3.42 PEN.

Values are per million tokens, before taxes and fees. For Argentina the official reference rate is used; if you pay at another rate, replace it in the formula of section 3.

Go deeperAI API prices in September 2026: what every model costs

Documentation: docs.mistral.ai/models/ministral-3-14b-25-12 ↗ · docs.mistral.ai/models/mistral-small-4-0-26-03 ↗ · developers.openai.com/api/docs/pricing ↗ · docs.mistral.ai/models/codestral-25-08 ↗ · api-docs.deepseek.com/quick_start/pricing ↗ · docs.mistral.ai/models/mistral-large-3-25-12 ↗ · ai.google.dev/gemini-api/docs/models/gemini-3.8-flash ↗ · ai.google.dev/gemini-api/docs/pricing ↗ · platform.claude.com/docs/en/models/overview ↗ · docs.mistral.ai/models/mistral-medium-3-5-26-04 ↗ · docs.x.ai/developers/models/grok-4.6 ↗ · Codifly: open AI API price dataset (JSON) ↗ · ExchangeRate-API: reference exchange rates ↗

2. Worked example: 100,000 conversations a month#

Assumption: 100,000 conversations a month, each with 1,500 input and 500 output tokens (150 million in, 50 million out). Monthly cost = 150 × input price + 50 × output price, in dollars, converted at the rate in section 1.

In this sample the priciest model costs 30 times the cheapest in any currency: picking the right model for the task matters more than any exchange-rate negotiation.

3. How many tokens your budget buys#

The practical question is not what a million costs but how many tokens a fixed budget reaches. The table shows, for four monthly budgets, the millions of blended tokens (3:1) each model buys, at the section 1 rate and before taxes.

If your product needs 50 million tokens a month, 13 of the 17 models fit a budget of one million Colombian pesos.

4. What this table leaves out#

The list price is not what your bank or card charges. On top of the converted value you add, depending on your country and payment method, the taxes that apply to foreign digital services, your card or bank's fee over the exchange rate and, in some countries, withholdings. There is no single figure: check your tax authority's current rules and your card statement.

local_cost = price_USD × exchange_rate × (1 + surcharge)
# surcharge = taxes + bank fee, as a fraction (0.10 = 10%)

Surcharges in this article are a sensitivity parameter, not any country's official rate.

5. Reproduce the numbers with your own data#

The dataset and the exchange rate are public. This command downloads both and computes the blended price of the five cheapest models in Colombian pesos.

curl -s https://codifly.co/data/ai-models.json > /tmp/m.json
RATE=$(curl -s https://open.er-api.com/v6/latest/USD | jq .rates.COP)
jq -r --argjson r "$RATE" '.models | sort_by(.blended_usd_per_mtok) | .[:5][]
  | "\(.name)\t\(.blended_usd_per_mtok * $r)"' /tmp/m.json

• Codifly open dataset (JSON, CC BY 4.0): data/ai-models.json.
• Exchange rates: ExchangeRate-API (daily reference, with attribution).

Documentation: Codifly: open AI API price dataset (JSON) ↗ · ExchangeRate-API: reference exchange rates ↗ · Creative Commons: CC BY 4.0 license ↗