AI model API prices in October 2026: Claude, GPT, Gemini, DeepSeek

AI model API prices are scattered across vendor sites, quoted in different units, and change more often than articles about them get updated. Below is a summary table as of October 7, 2026: the official vendor price and the Zukko price for input, output and cache for Claude, GPT, Gemini, DeepSeek, Qwen and GLM, an explanation of how to read it, and a calculation for a typical monthly volume.

How to read the prices

• The unit is US dollars per 1 million tokens. A token is a piece of text; Russian text usually takes more tokens than the same meaning in English.
• Input is everything you send: instructions, history, documents. Output is what the model writes. For most models, output costs more.
• Cache. If the start of a request repeats, the provider can serve it from cache. Cache reads are billed at their own price. Cache writes are billed separately only for Claude, and only in the Anthropic format.
• Official vendor price is the price when buying directly from the model developer. Zukko price is what's deducted from your balance for a request through the gateway.
• Context is how many tokens the model handles in one request.

AI model API prices: table as of October 7, 2026

All prices in USD per 1M tokens.

Claude

GPT

Gemini, DeepSeek, Qwen, GLM

The catalog also has previous Claude versions; their prices are on the Models page. For the current list and prices, call GET https://api.zukko.pro/v1/models (see the docs↗) or open the Models page in the dashboard.

What matters about caching

• Claude — the cache discount applies only in the Anthropic format: /v1/messages, Claude Code, Anthropic SDK. There both reads and writes are counted. In the OpenAI format (/v1/chat/completions) all Claude input is billed at the input price. How to choose a format is in the docs↗.
• GPT — only cache reads are visible (prompt_tokens_details.cached_tokens), in both formats. The first request is billed entirely at the input price.
• DeepSeek, Qwen, GLM, Gemini — reads are counted if the provider reports them in the response; writes aren't billed. For DeepSeek, Qwen and GLM the cache read price in Zukko equals the input price, so caching doesn't affect their bill.

The "Vendor" columns show only input and output prices. Vendors have their own cache prices, and when comparing heavy-cache scenarios you need to take those from the vendor's site — otherwise the comparison isn't fair.

Calculation for a typical volume

Scenario 1: 10M input and 2M output tokens per month, no cache

For example, a support bot or document processing. Prices are per 1M, so the formula is 10 × input price + 2 × output price.

Scenario 2: an agent with a large repeated context

50M input tokens per month, of which 45M are read from cache, 3M are written to cache and 2M are fresh input; plus 1M output. Zukko prices only, since vendor cache prices aren't in our table:

• Claude Sonnet 5, Anthropic format: 2 × 0.4 + 45 × 0.05 + 3 × 0.5 + 1 × 2 = 0.8 + 2.25 + 1.5 + 2 = $6.55;
• Claude Sonnet 5, OpenAI format (no cache discount): 50 × 0.4 + 1 × 2 = $22;
• GPT-6.1 Sol (writes not billed, fresh input 5M): 5 × 0.27 + 45 × 0.013 + 1 × 1.32 = 1.35 + 0.585 + 1.32 = $3.255.

The gap between the first two lines is the cost of the connection format you choose for Claude, not of the model.

All totals are example calculations on October 7, 2026 prices. Your result depends on the input/output ratio, the cache share and conversation length; see the real token counts in the usage field of each response. How price relates to the models' intelligence index is shown in the AI model ranking for October 2026↗.

Prices change

Providers revise prices often: for example, on October 6, 2026 Claude became roughly twice as expensive. So:

• check prices before any large budget calculation — in the catalog or via the API;
• don't build decisions that only work at today's price ratio between two models;
• set daily and monthly dollar limits on your keys so a price change doesn't come as a surprise (how to set them up↗).

You can also get the current model list from code:

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.zukko.pro/v1", api_key=os.environ["OPENAI_API_KEY"])

for model in client.models.list():
    print(model.id)

The Zukko balance is kept in dollars. You can top up via SBP or a Russian card in rubles at the market rate at the time of payment, via cryptocurrency, or with a foreign card. If you want to run your own scenario on real tokens, create a key in the dashboard↗ — one key works with every model in the table.

FAQ

Which model is the cheapest right now?

At October 7, 2026 prices, DeepSeek V4.1 Flash: $0.013 per 1M input and $0.038 per 1M output tokens. Cheapest doesn't mean suitable: test the quality on your own tasks.

Why does Claude have a "cache write" price and the others don't?

The gateway sees cache writes only for Claude in the Anthropic format and bills them at a separate price. For GPT and the other models, writes aren't billed separately.

Prices are in dollars — how much is that in rubles?

The balance is kept in dollars, and when you top up via SBP or a Russian card, rubles are converted at the market rate at the time of payment.

How often is this table updated?

The table is fixed as of October 7, 2026. Current prices are always on the Models page in the dashboard and in the GET /v1/models response.