$1 Budget: One Python Script, Two AI Models

01 / Start with a $1 learning envelope, not a subscription

You do not need separate Python integrations to try two AI models. An OpenAI-compatible relay lets the same client send requests to different models by changing the model field. Keep the prompt identical, compare the answers, and stop before turning a small experiment into a batch job.

Use $1 as your personal learning envelope, not as a promised price or a stated minimum deposit. CheapRelay offers pay-as-you-go access with no monthly fee or subscription. Billing follows actual tokens, including cache reads and writes. Check current model rates and available funding options before adding money.

Sign up for CheapRelay, create an API key, and copy the exact API model IDs for two models you want to test. DeepSeek V4 Flash and Gemini 2.5 Flash are two names to look for; their display names are not necessarily the strings your script must send.

02 / Reserve $0.10 for the first two-request test

Within that hypothetical $1 envelope, earmark at most $0.10 for your first experiment. This is a planning allowance, not an estimate of what two requests cost. If a conservative estimate using the listed rates exceeds it, shorten the request or choose another available model.

For ordinary uncached text, estimated cost equals input tokens multiplied by the input rate, plus output tokens multiplied by the output rate, with both rates converted from their published units. If rates are quoted per million tokens, divide each token count by 1,000,000 first. Cache charges and any separately billed token categories require their own calculation.

A short prompt and a requested output limit help control usage, but neither creates a hard dollar cap. Use an account spending limit if one is available. Otherwise, check your balance or usage after each small test.

03 / Install one SDK and keep the key out of the file

Create a working folder and, preferably, a Python virtual environment. Install the client with this terminal command:

python -m pip install openai

Create a file named compare_models.py. Copy the following Python paragraphs into it in order, each as its own line. The script asks for your key without displaying it, so you do not need to paste a secret into source code.

from getpass import getpass

from openai import OpenAI

client = OpenAI(api_key=getpass("API key: "), base_url="https://api.cheapapirelay.com/v1", timeout=30.0, max_retries=0)

first_model = input("First exact model ID: ").strip()

second_model = input("Second exact model ID: ").strip()

messages = [{"role": "user", "content": "Explain a Python list in three short sentences for a beginner."}]

first = client.chat.completions.create(model=first_model, messages=messages, max_tokens=120)

print("FIRST:", first.choices[0].message.content, "USAGE:", first.usage)

input("Check billed usage before continuing; press Enter to call the second model, or Ctrl+C to stop: ")

second = client.chat.completions.create(model=second_model, messages=messages, max_tokens=120)

print("SECOND:", second.choices[0].message.content, "USAGE:", second.usage)

Run it with python compare_models.py. Both calls use the same key and base URL. Only the model ID changes. The pause gives you a chance to inspect the first charge before authorizing another request.

04 / Check compatibility before spending on repeated calls

This example uses the OpenAI-compatible route under /v1, not the Gemini-native route under /v1beta. Confirm that each selected model supports chat completions and the max_tokens parameter. Parameter support can differ; use the documented equivalent if necessary rather than removing limits blindly.

The 120-token setting requests a small output allowance; it is not a billing guarantee. Reasoning, caching, and usage reporting can vary by model. Compare the returned usage with the billing record rather than treating visible answer length as the complete bill.

Automatic SDK retries are disabled here to avoid silently repeating requests. A timeout does not prove that the provider performed no work or charged nothing. Inspect usage before rerunning a failed call.

05 / Keep the cheaper adequate answer, not another recurring expense

Judge both answers on correctness, clarity, and whether they followed the three-sentence instruction. Record the billed cost beside that judgment. Two answers are a learning exercise, not a benchmark.

For your next script, make one model the default and call another only when needed. Keep prompts short, avoid parallel batches while learning, and review usage regularly. One key simplifies access; disciplined request counts protect the budget.