![translation](https://cdn.durumis.com/common/trans.png)
This is an AI translated post.
Gemini 1.5 Flash, GPT-4o, and Pricing of Other LLMs
- Writing language: Korean
- •
-
Base country: All countries
- •
- Information Technology
Select Language
Summarized by durumis AI
- Compare the performance and pricing of various AI models such as GPT-4o, Opus, Gemini 1.5 Pro, Haiku, and Gemini 1.5 Flash to present the strengths and weaknesses of each model, and recommend appropriate models based on the purpose of use.
- Provides users with the best AI model selection guide considering input token size, output ratio, and task complexity.
- Based on the performance and pricing information of the latest AI models as of May 30, 2024, this helps users make wise choices.
Google and OpenAI have both announced a ton of new AI-related content in the past two days.
The two things people are generally curious about are:
Performance and price. (Of course there are more features, but that's something professional bloggers will review...)
Open AI - GPT
As always with OpenAI, the new 4o is cheaper than the previous GPT-4T. Performance is something you can find a ton of reviews about on other blogs, so let's just talk about the price here.
GPT Price Table
Open AI has consistently lowered the price each time a new product is released since GPT 3.5 Turbo, starting with the initial release of GPT 4. Of course, the performance has been upgraded. Currently, if you're looking for the most affordable option, 3.5 Turbo is the way to go. For everything else, 4o should do the trick.
Anthopic - Claude 3
Claude 3 Price Table
Anthropic hasn't released any new products recently, but Haiku for value and Opus for high performance make it an LLM company that can't be ignored.
Haiku is the cheapest of the three in terms of input token pricing, making it the most affordable for simple text processing.
In fact, until Gemini Flash came out, Haiku even outperformed Gemini 1.0 Pro, making it a very useful LLM.
Google - Gemini
Gemini Price Table
Google maintains two pricing systems.
One is AI Studio, the other is Vertex AI.
AI Studio is token-based like other companies, while Vertex AI is uniquely priced based on characters.
Looking at the table above, if 1 token is less than 3 characters on average (1-2 characters), Vertex AI is cheaper to use. If it's 3 characters or more, AI Studio is cheaper. But, naturally, English characters usually exceed the character count, making AI Studio cheaper. Korean nowadays also has cases where 1 token is multiple characters...
Anyway, even if you only look at input tokens or performance, Gemini 1.5 Flash is much better than 1.0 Pro. For high-performance tasks, 1.5 Pro is superior.
Summary
Comprehensive
In terms of performance alone, based on MMLU, it seems to be GPT-4o > Opus > 1.5 Pro.
For highly intellectual tasks, use GPT-4o. If you want something a little cheaper or if your token size exceeds 200K (Opus only supports up to 200K), Gemini 1.5 Pro might be a good option. It's a bit different in actual usage, so use what works best for you.
If you need to do a lot of text processing on a budget, there are two options:
If the ratio of input to output is low (e.g., if you need to input a large amount of documents and output a short result), Claude 3 Haiku is the cheapest. However, Haiku has a high output cost, so on the other hand, if the ratio of output is higher (e.g., if you input a specific text and then instruct it to modify or change it), Gemini 1.5 Flash is recommended. In that case, Flash is the cheapest for output cost.
Summary and Conclusion
"I don't care about the price, I just want the most complex task done." -> GPT - 4o
"But, the input token size exceeds 128K." (GPT - 4o only supports up to 128K) -> Opus
"High performance is needed, but the price is a bit cheaper, or the token size exceeds 200K." (Opus only supports up to 200K) -> Gemini 1.5 Pro
"I need the cheapest LLM." -> Haiku
"But, the input/output ratio is higher in terms of output, or it exceeds 200K tokens." -> Gemini 1.5 Flash