If you're looking for a cheap ChatGPT API, start by separating two products: a ChatGPT subscription and the OpenAI API. ChatGPT Plus is a $20/month app subscription with usage limits; the API is a pay-per-token developer service. Codex4Less provides the second. Its promise is straightforward: the same Codex models, up to 97% cheaper.

This guide tests that promise against official OpenAI list prices for three GPT models: GPT-6 Astra, GPT-6.1 Sol and GPT-5.6 Sol. It calculates a realistic monthly bill for 10 million input and 2 million output tokens, then shows where the savings come from — and what to check before you move a production workload to a cheaper API.

The short answer: three models, side by side

ModelOpenAI (input / cache / output)Codex4Less (input / cache / output)Unit-price difference
GPT-6 Astra$10.00 / $1.00 / $50.00$1.20 / $0.10 / $5.0088–90%
GPT-6.1 Sol$2.00 / $0.10 / $10.00$0.10 / $0.02 / $1.0090–95% (cache 80%)
GPT-5.6 Sol$4.00 / $0.40 / $20.00$1.50 / $0.20 / $1.5062.5–92.5% (cache 50%)

Prices are USD per 1 million tokens and were verified on October 21, 2026. OpenAI prices are the Standard tier for prompts up to 272K tokens. Above 272K, OpenAI charges 2x for input and cached input and 1.5x for output. Check Codex4Less's current pricing page for its long-context terms.

"Cache" means cached input tokens from prompt caching. The discount is the unit-price difference on that line item; total savings depend on your cache hit rate. OpenAI's promotional pricing for GPT-5.6 Sol is available at least through November 21, 2026, so recheck the table after that date.

Codex4Less's "up to 97%" claim is the catalog-wide ceiling; for the three GPT models in this guide, the unit-price difference ranges from 62.5% to 95%.

Cheap ChatGPT: subscription or API?

Most people searching for a cheap ChatGPT plan are really looking for a lower-priced ChatGPT subscription. The honest answer: ChatGPT Plus is $20/month, and Pro plans start at $100/month. Codex4Less does not sell ChatGPT subscriptions. It sells OpenAI-compatible API access, so it does not replace Plus feature-for-feature.

What you needBetter fitWhy
A few chats, images and research tasks per dayChatGPT Plus ($20/month)Fixed price, ready-made interface, included features
Heavy Codex CLI / IDE agent workAPI (pay per token)Pay for usage, choose models, automate workflows
Your own app or background jobsAPINo interface required, scales with demand
Chat UI plus API accessOpenAI-compatible client + APIKeep the interface and billing as separate layers

Codex is included in ChatGPT plans with plan-specific usage limits. When you use it with an API key, you pay per token instead. That distinction matters: "cheap Codex" can mean reducing a subscription bill or reducing API token spend — and the right fix is different for each.

Be careful with shared accounts and "unlimited ChatGPT" offers. Account access, privacy and service continuity are all at risk, and resold accounts outside OpenAI's official plans cannot be reliably verified. On the API side, the basic safety checks are simple: the key belongs to you, usage is visible in a dashboard, and the model ID can be verified.

What is Codex API — and how do you make Codex cheaper?

Codex API is API access to the models that Codex CLI or an IDE extension uses in the background. It is not a separate model name. In a Codex workflow, you use models such as GPT-6 Astra, GPT-6.1 Sol or GPT-5.6 Sol. Codex CLI can run with a ChatGPT account (within plan limits) or with your own API key (pay per token).

The main cost drivers in Codex are:

  • Input tokens: Repository context, system prompts, tool output and previous turns.
  • Output tokens: Generated code, explanations and plans.
  • Cached input: Repeated context read at a lower price.
  • Agent loops: Every tool call and retry generates more tokens.
  • Context length: Above 272K tokens, OpenAI raises unit prices.
  • Model choice: Astra, 6.1 Sol and 5.6 Sol have different unit costs.

A cheap Codex setup is not only about a lower unit price. It is also about wasting fewer tokens. Completing the same task with a tighter context can matter as much as the price gap:

  • Select relevant files instead of pasting the whole repository.
  • Keep system prompts and repeated context stable so caching can work.
  • Ask for a patch or diff instead of a full-file rewrite.
  • Match reasoning effort to the task; the highest setting is not always necessary.
  • Watch agent loops and failed retries — they multiply cost quickly.
  • Use Astra for the hardest work and 6.1 Sol for everyday coding.

Which model should you use?

ModelBest forCodex4Less input / output
GPT-6 AstraHardest reasoning, large refactors, agents that need high accuracy$1.20 / $5.00
GPT-6.1 SolNear-Astra performance, heavy daily coding, cost/performance balance$0.10 / $1.00
GPT-5.6 SolGPT-5.6 family flagship, existing GPT-5.6 workflows$1.50 / $1.50

OpenAI positions Astra as its most capable model, 6.1 Sol as "near-Astra performance at lower cost", and 5.6 Sol as the flagship of the GPT-5.6 family. That does not guarantee identical results on your workload. Run a small A/B test on real tasks; if quality is close, move the routine work to the cheaper model.

For the results of our four-task mini test and task-by-task recommendations, see GPT-6 Astra vs GPT-6.1 Sol vs GPT-5.6 Sol.

Monthly cost example: 10M input + 2M output

Use this formula:

Monthly cost = (input tokens / 1,000,000 × input price) + (output tokens / 1,000,000 × output price)

ModelOpenAICodex4LessMonthly saving
GPT-6 Astra$200.00$22.00$178.00 (89%)
GPT-6.1 Sol$40.00$3.00$37.00 (92.5%)
GPT-5.6 Sol$80.00$18.00$62.00 (77.5%)

This calculation assumes no cached input. Codex-style agents often reuse a large part of their context, which lowers the official bill too. For example, if 80% of Astra's 10M input tokens are cache hits, the official cost falls to $128.00 and the Codex4Less cost to $13.20 — still roughly 90% savings. Calculate your own cache ratio from your usage records.

Token counts are illustrative. Real cost depends on model choice, context length, cache hit rate, output length and agent behavior.

7 checks before choosing a cheap ChatGPT API

  1. Model identity: Ask for the exact ID (gpt-6-astra, gpt-6.1-sol) rather than a generic "GPT-6" label. The provider's GET /v1/models response should list it. A different version or a throttled pool can erase the advertised advantage.
  2. Price components: Are input, cached input and output priced separately? Are long context, image input or tool calls extra? What are the minimum balance, refund and withdrawal terms?
  3. OpenAI compatibility: Does your existing OpenAI SDK work after changing only base_url and the API key? Are /v1/chat/completions, /v1/responses and streaming supported? This directly affects integration cost.
  4. Billing transparency: Can you see the model, token count and cost per request? Run the same prompt through both providers and compare the returned token counts. Do not rely on the advertised discount alone.
  5. Data policy: If you send code, customer data or personal data, read the retention policy, training policy and deletion process. For agencies and production teams, this matters more than a few cents per million tokens.
  6. Limits and error handling: Are rate limits, 429 responses, timeouts and support channels clear? Do not run production without retry/backoff and a budget alert.
  7. Too-cheap claims: "Unlimited" or "shared account" offers carry model-quality, privacy and continuity risks. You should own the API key, see usage in a dashboard and be able to verify the model ID.

How to use Codex4Less as your cheap ChatGPT API

Creating a Codex4Less account is free. According to the pricing page, balances start at $5 and never expire; you need to top up before sending requests.

  1. Create an account. Sign up with your email at codex4less.com/register.
  2. Top up your balance. Load only what you need; there is no monthly commitment.
  3. Create an API key. The key is shown only once at creation. Store it securely.
  4. Change the base URL. In your OpenAI SDK, set base_url to https://api.codex4less.com/v1 and choose a model ID.
python
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.codex4less.com/v1",
    api_key=os.environ["CODEX4LESS_API_KEY"],
)

response = client.chat.completions.create(
    model="gpt-6-astra",
    messages=[
        {"role": "user", "content": "Give me a short sample response."}
    ],
)

print(response.choices[0].message.content)

For Codex CLI, set wire_api = "responses" in ~/.codex/config.toml. The full setup is in the Codex CLI guide.

Never hard-code an API key in production. Use an environment variable or a secrets manager, and keep keys out of screenshots, chat messages and repositories.

FAQ

Is a cheap ChatGPT API safe to use?

It depends on the provider. Verify the model ID, data policy, billing transparency and rate limits. Codex4Less is an independent platform and is not affiliated with OpenAI. If you send sensitive code or personal data, review its privacy and retention terms separately.

Does Codex4Less sell ChatGPT Plus?

No. Codex4Less provides API access, not a ChatGPT app subscription. If you need a chat interface, you can connect an OpenAI-compatible client with your own API key, but Plus features such as deep research and image generation are not identical.

What is Codex API?

It is API access to the models used by Codex CLI or an IDE extension. Codex CLI can run with a ChatGPT plan or an API key; the API-key path is billed per token. Codex4Less provides an OpenAI-compatible endpoint for Codex CLI.

Why are prices lower than OpenAI's official rates?

Codex4Less operates a balance-based API gateway and sets its prices below OpenAI's list prices. The site states that models run as original models on official provider infrastructure. You can test that claim by running the same prompt through both providers and comparing response quality and token counts.

GPT-6 Astra or GPT-6.1 Sol?

Use Astra for the hardest reasoning and coding tasks; use 6.1 Sol when cost/performance matters. OpenAI describes 6.1 Sol as near-Astra performance at a lower price. Run an A/B test on your own tasks before switching a production workflow.

Which costs should I compare when looking for a cheap ChatGPT API?

Compare input, cached input and output prices separately. Factor in your monthly input/output volume, cache hit rate and whether you use contexts above 272K tokens. Looking only at the input price can be misleading.

How does billing work?

You load a balance, and input, cache and output tokens are deducted from it. According to Codex4Less's pricing page, balances start at $5 and never expire. Check the pricing page for current terms and payment methods.

Is there a free trial?

Creating an account is free; usage requires a balance. The site does not advertise free credits. Start with a small top-up and short prompts to keep the first test inexpensive.

Conclusion

The decision is not just about the advertised discount percentage. It is about which model you use, how many input and output tokens you send, and how much of that context is cached. GPT-6 Astra fits the hardest work, GPT-6.1 Sol fits heavy daily coding, and GPT-5.6 Sol fits existing GPT-5.6 workflows. Across all three, Codex4Less prices sit below OpenAI's Standard list; in the 10M input + 2M output scenario, savings range from 77.5% to 92.5%.

Compare the numbers with your own token volume on the pricing page, then run a small test with a small balance. Create your Codex4Less account, generate an API key and test the same prompt on your real workload.

Compare prices → Create a free account →

Sources

Related guides

Codex4Less is an independent platform and is not affiliated with OpenAI. GPT and Codex are trademarks of OpenAI.