Save Money on Massive Outputs: Alternatives to Consider Beyond OpenAI

Choosing the right AI provider can save you time, money and headaches. While OpenAI's GPT-4o Mini is known for its impressive capabilities (including vision), there are many other platforms - such as Together.ai or Fireworks.AI - with comparable pricing structures. Which option is right for you often depends on how many tokens you feed (input) versus how many tokens the model generates (output).

Understanding Input vs. Output

• Input Tokens: The text you send to the model (questions, instructions, data).
• Output Tokens: The text the model returns as a response.

Most AI platforms charge you separately for input and output tokens. When your usage is asymmetric—that is, if you’re generating a lot more output than you’re supplying as input—some services could end up costing significantly more than others.

Cost Comparison Example (10k Input / 5k Output)

Let’s compare OpenAI’s GPT-4o Mini to Together.ai’s 8B model using a hypothetical job (e.g. summarizing data from the Web) with 10,000 input tokens and 5,000 output tokens:

This cost calculation is based on the current prices in Openai and Together AI as of 12/24.

OpenAI GPT-4o Mini

- Input: 10,000 tokens × $0.150 per 1M tokens = $0.0015
- Output: 5,000 tokens × $0.600 per 1M tokens = $0.003
- Total: $0.0015 + $0.003 = $0.0045

Together.ai 8B

- Input: 10,000 tokens × $0.10 per 1M tokens = $0.001
- Output: 5,000 tokens × $0.18 per 1M tokens = $0.0009
- Total: $0.001 + $0.0009 = $0.0019

In this scenario, Together.ai costs less than half of OpenAI’s GPT-4o Mini for the same number of tokens.

When OpenAI Might Be Cheaper

Despite Together.ai’s lower cost here, OpenAI can still be more economical if your input is much larger than your output. Suppose you’re primarily sending large documents for analysis and only requesting brief summaries in return—OpenAI’s pricing might actually come out ahead. Additionally, OpenAI’s infrastructure is often praised for speed and reliability, which can be crucial for certain use cases.

When to Consider Alternatives

If your work involves significantly more output than input (for instance, generating lengthy articles, stories, or large data sets), platforms like Together.ai or Fireworks.AI could save you a factor of four or even more. These solutions often specialize in text-heavy tasks and can provide cost benefits without sacrificing too much in terms of accuracy or speed, especially if you don’t need extra features like vision.

Wrap-Up

When it comes to choosing a model, understanding the task at hand is key.

• Evaluate Your Token Balance: Understand if your tasks are input-heavy or output-heavy.
• Compare Costs: Don’t just assume OpenAI is the cheapest—other platforms might surprise you.
• Check Performance: Cheaper doesn’t always mean better, so be sure to test each service for the specific tasks you need.

Ultimately, there’s no one-size-fits-all answer. By weighing how many tokens you’ll send in versus how many you’ll generate, you can pick the provider that fits your budget and performance needs—whether that’s OpenAI or an alternative like Together.ai.