Free ChatGPT and cheap AI in Uzbekistan, 2026

Free ChatGPT, Gemini and Claude in Uzbekistan, plus dirt-cheap Chinese models: DeepSeek and OpenRouter prices, Harness, Mars Code. As of November 2026.

Free ChatGPT and cheap AI in Uzbekistan, 2026

In short, as of 3 November 2026: a free ChatGPT tier exists, and Uzbekistan is on the official list of supported countries. Gemini, Claude and Microsoft Copilot also have free tiers. What they include and the limits change over time, so check the plan page on the day you sign up.

Western services can be used free in the chat. Chinese models are usually not free, but cost pennies through the API, and below is exactly how much. They have a downside too: your data goes abroad. Paid plans and payment are covered in a separate article: ChatGPT and Claude in Uzbekistan.

What is free and works from Uzbekistan

  • ChatGPT. A free tier exists and Uzbekistan is on the official country list. We could not verify the exact Free limits: OpenAI pages did not open during our check.
  • Gemini. Uzbekistan and the Uzbek language are on the app's official list. Free costs 0 dollars, needs a Google account, and includes the Flash-Lite model and image generation. The limit is not given as a number: it refreshes every 5 hours up to a weekly cap.
  • Claude. A Free plan exists and Uzbekistan is on Anthropic's list of supported countries. It covers web, desktop and phone, web search and voice mode; usage is limited, with no numbers given. Claude Code is not on Free.
  • Microsoft Copilot. Free with a Microsoft account, some features limited. The page has no country list and does not name Uzbekistan, but Uzbek and Russian are among the interface languages. Open the service to see whether it works for you.
  • Meta AI. We found no official confirmation that it is available in Uzbekistan, which does not mean it is unavailable. Check in WhatsApp or Instagram.

Chinese models: not free, but cheap

Chinese models such as DeepSeek, Qwen, Kimi and MiniMax are strong as a cheap API, not as a free chat. OpenRouter lists about fifteen free models with the :free suffix, and almost none are Chinese: NVIDIA, Google Gemma, Cohere and others. The Chinese ones there are paid but inexpensive: the cheapest start at a few cents per 1M input tokens, the pricier ones cost dollars, and the price depends on the provider.

DeepSeek directly. Official off-peak prices per 1M tokens: deepseek-flash charges $0.15 for input and $0.6 for output, with cached input at $0.003 off-peak; deepseek-v4-pro charges $0.66 and $1.98. Peak-hour prices are double, and the context is 1M tokens. Peak on weekdays is 06:00-09:00 and 11:00-15:00 Tashkent time; weekends are fully off-peak.

What that means in money. A request with 2000 input tokens and a 1500-token answer (about 1000 words) on deepseek-flash costs about $0.0012 off-peak. Five dollars covers roughly 4000 such requests, half that at peak. This is a rough estimate: it ignores caching and paid Search and Fetch.

An agent on top of the key. DeepSeek Harness starts with the dsh command and opens a web interface: you enter a DeepSeek API key in Settings, Models, choose a project folder, and the agent reads and edits files, runs commands and keeps a plan. For operations needing approval the interface asks first, under the active policy. The DeepSeek API also understands the Anthropic format (base URL api.deepseek.com/anthropic), so people connect it to Claude Code too. Search and fetch cost $0.75 per 1000 requests, billed in Harness as well from 1 November 2026.

Calling it through OpenRouter. You need to register and get an API key: requests go to openrouter.ai/api/v1/chat/completions with an Authorization header, and the model is written as "developer/model". Free :free models are limited: 20 requests per minute and 50 per day without topping up, up to 1000 per day after buying about 10 dollars of credits. The current free list is on the OpenRouter page.

On data. Under DeepSeek's policy, data is stored in China, your text, files, history and IP are collected, and model training is among the stated purposes. Opting out is done by emailing privacy@deepseek.com. OpenRouter itself does not log requests by default, but some model providers may use inputs for training. Do not send secrets or other people's personal data to any service on this list.

Mars Code: AI for kids and teenagers

If you are choosing AI for a child, we have our own option. Mars Code runs in the browser: the child describes an idea and the AI builds a game, site or app. It is designed for ages 7-17.

Sign-in is only through Mars ID; this is not an open public chat. Models go through our Mars Token Factory gateway, which has a free pool that routes a request to the best live free model. When coins on a paid model run out, the request gets an answer from that pool instead of a refusal.

This is a pilot. The list of free models changes, and we do not promise speed or limits.

When not to rush

  • Data in chats. Do not paste contracts, passwords, or children's passport and school data into a public chat. This matters most with DeepSeek, since data is stored in China.
  • Age. Anthropic's terms are aimed at users aged 18 and over. DeepSeek states the service is primarily for adults.
  • Limits. Free limits refresh every 5 hours or daily and can cut work off midway.
  • Prices. API prices change: check the pricing page before topping up.

What to do today

  1. Sign up for ChatGPT or Gemini and solve two or three real tasks: a letter, a table, a translation.
  2. Compare the answer across two services: there is no single best AI, it depends on the task.
  3. If you need AI for a child, open Mars Code and sign in with Mars ID.

Sources