Chatbot API Free: Options, Limits and Trial Credit

The four kinds of free

When people ask for a chatbot api free of charge, they want to build or test a bot without paying up front. That is reasonable, and there are four ways to do it. They differ in what you give up.

Kind	What you get	What you give up
Provider free tier	A monthly or daily quota of requests	Low rate limits, quotas that change, sometimes lower-grade models, data terms to read
Open-weight model run locally	No per-token fee, full control	Hardware, setup, slower generation, you host the endpoint
Demo or community endpoint	A quick way to try a chat	No stability, no guarantee, not meant for an app
Trial credit on a paid API	A real production-style endpoint	A small amount, and a time limit

None of these is a permanent free plan for a bot with users. They are ways to build and test. The trick is to pick the one whose limits you can live with while you prove the idea.

Free tiers: what to check

Several providers publish free tiers for their chat models. I will not quote any of them, since quotas change. When you read one, check:

• Requests per minute and per day. A bot with ten simultaneous users can hit a low cap quickly.
• Which models are included. Free often means the smaller model.
• Data use. Some free tiers allow the provider to use prompts for improvement. Read the terms before sending anything private.
• Regional availability and sign-up requirements.
• What happens at the limit. A 429 error, a slowdown, or a forced upgrade.

A free tier is excellent for a prototype and a weekend project. For a launched bot, price the paid tier before you build around the free one.

Self-hosting an open-weight model

Open-weight chat models can be run with tools like llama.cpp or Ollama, which expose an OpenAI-style endpoint on your machine. The software is free. The costs are the GPU or enough RAM to hold the model, electricity, and your time. A 4-bit 8B model needs roughly 4 GB for weights, so many laptops can run it slowly. Serving several users at once is where a desktop starts to struggle, since a single GPU usually handles one request at a time at full speed.

This suits private experiments and offline demos. It suits a public bot less, because uptime and scaling become your job.

Trial credit: a real endpoint for a short test

New users here can claim $0.50 of trial credit with one click on the API key page after signing up with Google or with email and password. It needs no card, it is one per person, and it is valid for 7 days. A trial key allows 2 concurrent requests and 60 requests per minute, so it is enough to test a bot with a handful of simultaneous users.

The API is OpenAI-compatible, so any OpenAI SDK works after you change the base URL and key. The model ID is uncensored. It is one open-weight model run by an independent provider, tuned to answer without content refusals for lawful adult use. It supports streaming, tool calling and JSON mode, has a 100,000 token context window, and is text only. It is not the official API of any model vendor.

What $0.50 buys

Prices are $0.15 per 1M input tokens and $0.50 per 1M output tokens. Failed or refused requests are not charged. Two examples:

Short one-off requests

A request with 1,000 input and 500 output tokens costs 1,000 x $0.15 / 1M = $0.00015 for input and 500 x $0.50 / 1M = $0.00025 for output, so $0.0004. Then $0.50 / $0.0004 = 1,250 requests.

Ten-turn conversations

A chatbot resends history each turn, so cost grows faster than turn count. Assume a 300-token system prompt, 150 tokens per user message and 250 per reply. Input at turn n is 300 + 150 + 400 x (n - 1). Over ten turns that is 10 x 450 + 400 x 45 = 22,500 input tokens, and 10 x 250 = 2,500 output tokens.

• Input: 22,500 x $0.15 / 1M = $0.003375
• Output: 2,500 x $0.50 / 1M = $0.00125
• Per conversation: $0.004625
• $0.50 / $0.004625 = about 108 conversations

That is enough to put a bot in front of a few friends for a week.

A minimal chatbot you can run now

Here is a complete command-line chat loop with memory. Install the SDK with pip install openai, put your key in the file and run it.

Copyfrom openai import OpenAI

client = OpenAI(base_url="https://api.chatbotapis.com/v1", api_key="YOUR_KEY")

history = [{"role": "system", "content": "You are a friendly assistant. Keep answers short."}]

while True:
    user = input("you> ")
    if not user:
        break
    history.append({"role": "user", "content": user})
    reply = client.chat.completions.create(model="uncensored", messages=history)
    text = reply.choices[0].message.content
    history.append({"role": "assistant", "content": text})
    print("bot>", text)

To watch the credit drain, read the X-Balance-USD response header, or call GET /v1/balance. The docs cover streaming, tools and limits in full.

Stretching a small budget

A few habits make trial credit or any small balance last longer.

• Trim history. Since input grows with every turn, keep only the system prompt and the last several messages. Cutting the ten-turn example to the last four exchanges saves a large share of the 22,500 input tokens.
• Keep the system prompt short. It is billed on every request.
• Cap reply length. The optional max_tokens parameter limits output, which costs more per token than input.
• Answer repeated questions from a cache. If a bot gets the same ten questions all day, store the replies.
• Limit each user. A per-user message cap stops one visitor from draining the balance.
• Handle 429 politely. If you receive HTTP 429, wait and retry. An overload response includes a Retry-After: 5 header, and the error codes rate_limited and concurrency tell you which limit you hit.

CopyMAX_MESSAGES = 8  # keep the last 8 messages plus the system prompt

def trimmed(history):
    return history[:1] + history[1:][-MAX_MESSAGES:]

Pass trimmed(history) as messages in the loop above. Test that the bot still makes sense after trimming, and if it forgets important facts, put those into the system prompt.

Planning for when free ends

The usual trap is building on a free tier and then discovering that the paid price changes the economics. Do the math early. Take your expected conversations per month, multiply by the per-conversation cost above, and compare. At $0.004625 per ten-turn conversation, 10,000 conversations a month cost $46.25. That is a number you can put in a budget. For a view of how this compares more generally, see the page on LLM API pricing.

Here billing is prepaid and pay as you go with no subscription. Top-ups are any whole amount from $20 to $500, with +5% from $50, +10% from $100 and +15% from $200. You can pay by card (Visa, Mastercard, Apple Pay, Google Pay) or crypto (USDT on TRC20, USDC on Base). Credit never expires, and prompts are not used for training. Details are on the pricing page.

More guides

• Build a Chatbot with an OpenAI-Compatible API
• GPT 5 API vs Uncensored Alternatives: Cost & Features