AI API prices in October 2026: what each model costs

1. October prices, model by model#

Language-model APIs charge per million tokens, with one rate for the tokens you send (input) and another, almost always higher, for the tokens the model generates (output). The table gathers the standard prices of 21 models, read on October 17, 2026 on each provider's official page, sorted from lowest to highest blended price. The blended price assumes 3 input tokens for every output token: (3 × input + output) / 4.

The spread is still enormous. The cheapest output costs $0.20 per million (Ministral 3 · 14B) and the most expensive, $50 (GPT-6 Astra and Claude Fable 5.1): 250 times more. On input the range runs from $0.10 (Claude Haiku 5.5 and GPT-6 Luna) to $10: 100 times. Claude Haiku 5.5 and GPT-6 Luna tie at $0.10 / $0.50, and both have a context of over one million tokens.

Three figures in the table come with conditions. Claude Haiku 5.5 shows the rate for prompts of up to 100,000 tokens. Gemini 4 Argon shows its rate valid through November 30. DeepSeek shows the peak-hours rate. The details are in section 4; the table only includes the standard price, without caching or batch discounts.

Documentation: OpenAI: API pricing ↗ · OpenAI: GPT-6.1 Sol ↗ · OpenAI: GPT-6 Luna ↗ · Anthropic: Claude Sonnet 5.5 ↗ · Anthropic: Claude Haiku 5.5 ↗ · Anthropic: Claude models and pricing ↗ · Google: Gemini API pricing ↗ · Google: Gemini 4 Argon ↗ · xAI: Grok 4.7 ↗ · DeepSeek: API pricing ↗ · Mistral: Mistral Large 3 ↗

2. What changed since September#

The September guide covered 17 models. In recent weeks six new models appeared that change how prices are distributed, especially in the $2 input-per-million band, where five models from four providers now compete.

Two underlying moves. First, small models dropped to $0.10 input: Claude Haiku 5.5 and GPT-6 Luna cost half as much as GPT-5.6 Luna ($0.20 / $1.20), with a context of over one million tokens. Second, the mid band repeats itself: Sonnet 5.5, GPT-6.1 Sol and Gemini 4 Argon cost the same ($2 / $10), so the choice is no longer decided by list price but by your evaluation and the conditions in section 4. Claude Sonnet 5 and Claude Opus 5 remain active, but Anthropic marks them as legacy.

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

Documentation: Anthropic: Claude Sonnet 5.5 release ↗ · OpenAI: API changelog ↗ · xAI: release notes ↗

3. Worked example: a support bot for a month#

Suppose a support assistant that handles 100,000 conversations a month. Each conversation sends 1,500 input tokens (instructions, history and help articles) and receives 400 output tokens. That adds up to 150 million input tokens and 40 million output tokens. The calculation is direct: input tokens times the input rate, plus output tokens times the output rate.

PRICES = {  # USD per million tokens: (input, output)
    "Claude Fable 5.1": (10.00, 50.00),
    "Claude Opus 5.5": (4.00, 20.00),
    "Claude Sonnet 5.5": (2.00, 10.00),
    "Grok 4.7": (2.00, 6.00),
    "Mistral Small 4": (0.15, 0.60),
    "Claude Haiku 5.5": (0.10, 0.50),  # prompts up to 100,000 tokens
}

CONVERSATIONS = 100_000
INPUT, OUTPUT = 1_500, 400  # tokens per conversation

def monthly_cost(input_price: float, output_price: float) -> float:
    m_input = CONVERSATIONS * INPUT / 1_000_000
    m_output = CONVERSATIONS * OUTPUT / 1_000_000
    return m_input * input_price + m_output * output_price

for model, (input_price, output_price) in PRICES.items():
    print(f"{model}: {monthly_cost(input_price, output_price):,.2f}")

# Gemini 4 Argon changes rate on December 1, 2026
print("Argon through Nov 30:", f"{monthly_cost(2, 10):,.2f}")
print("Argon from Dec 1:", f"{monthly_cost(4, 20):,.2f}")

Output is only 21% of the tokens, yet it weighs 57% of the bill on Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, because on all of them it costs five times more than input. Grok 4.7, with a three-to-one ratio, brings it to 44%. Between the most expensive and the cheapest model in the example there is a factor of 100: from $3,500 to $35 a month.

The Gemini 4 Argon case shows why dates matter: at the current rate the example costs $700 a month; from December 1 it rises to $1,400. If you budget today for the full year, use the later rate, not the promotional one.

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

4. Conditions and dates that change the list price#

The table shows standard prices and excludes caching and batch discounts. These conditions can move your bill more than switching models, and several have an end date, so note when you verified each one.

• Rates with an end date: Gemini 4 Argon costs $2 / $10 through November 30, 2026 and $4 / $20 from December 1; Gemini 3.8 Flash costs $0.75 / $3.75 through December 31 and double from January 1, 2027; OpenAI keeps promotional pricing on GPT-5.6 Sol at least through November 21.
• Tiers by prompt size: Claude Haiku 5.5 charges $0.10 / $0.50 up to 100,000 tokens and $0.50 / $2.50 above; OpenAI models charge double input and 1.5 times output from 272,000 input tokens; Grok 4.7 moves to $4 / $12 from 200,000 tokens. The surcharge applies to the whole request.
• Batch and caching: OpenAI and Anthropic document a 50% discount for asynchronous requests through their batch API, and a cache read costs a fraction of standard input (10% on Claude Haiku 5.5 and 5% on Claude Sonnet 5.5 and Opus 5.5).
• Time of day: DeepSeek charges half outside its peak hours, 01:00 to 04:00 and 06:00 to 10:00 UTC Monday through Friday; on Chinese public holidays the whole day is off-peak.
• Retirements: Claude Haiku 4.5 is still active but its earliest retirement date was October 15, 2026; if you use it, plan the migration.

That is why it pays to store, next to each budget, the date you read the rate and the end date of any promotion. A budget built on today's prices can come up short in December.

Go deeperKimi K3, GLM-5.3, Qwen3.8 Max and Muse Spark pricing: the AI APIs rising on LLM Stats

Documentation: OpenAI: batch and cached input ↗ · Anthropic: prompt caching and Batch API ↗ · Google: Gemini 4 Argon and Gemini 3.8 Flash rates ↗ · xAI: 200,000-token threshold on Grok 4.7 ↗ · DeepSeek: peak and off-peak rates ↗ · Anthropic: model retirements ↗

5. How to choose by task#

Price is a filter, not the final criterion. A cheap model that fails and forces retries can end up costing more per solved task. The sensible order is to describe the shape of your workload, shortlist by price and context, and decide with an evaluation on your own cases.

For the shortlist, Codifly's price ranking sorts models by cost and the cost-benefit ranking crosses them with performance. The AI cost calculator does the section 3 math with your volume and your input-to-output ratio, and the AI model comparator shows the limits and conditions of each listing.

Go deeperHow to Choose an LLM for Production Without Relying on Leaderboards

6. Keep your numbers current#

AI API prices change faster than almost any other cloud rate: new models ship, older ones become legacy and promotions expire. Between the September guide and this one there were six launches.

The figures in this guide come from each provider's official pages, read on October 17, 2026. Codifly's open AI API price dataset is published as JSON and CSV under a CC BY 4.0 license: you can use it in your notebooks, articles or products with attribution. Compare its reviewed_at field with the date of your last budget, and remember it is a starting point, not a provider commitment.

curl -s https://codifly.co/data/ai-models.json \
  | jq -r '.reviewed_at, (.models | sort_by(.blended_usd_per_mtok) | .[:5][]
      | "\(.name)\t\(.input_usd_per_mtok)\t\(.output_usd_per_mtok)")'

• Automate a weekly query to the dataset and compare reviewed_at with the date of your last budget.
• Store billed tokens per model and per feature: without that you will not know which price change affects you.
• Before signing a usage commitment, open the provider's official page and confirm the rate and its end date.

Documentation: Codifly: open AI API price dataset (JSON) ↗ · Creative Commons: CC BY 4.0 license ↗