I just tried out the new GPT-4o mini model and am BEYOND Impressed
Synopsis
Despite how powerful GPT-4o is, I don’t tend to use it often. If I’m looking for a powerful language model for complex reasoning tasks, I’m picking up Claude 3.5 Sonnet 9/10 times. This is because of its marginally better performance with slightly better pricing.
But after today, this may no longer be the case.
Today, OpenAI released GPT-4o mini. Unlike the rest of the GPT-4 family, OpenAI made GPT-4o-mini to compete with smaller models like Claude Haiku. In their article, they claim that its one of the best language models for its size, and that its more than 60% cheaper.
And after playing with it for a couple hours, I wholeheartedly agree.
A July 2024 comparison of GPT-4o mini and GPT-3.5 Turbo
This July 2024 excerpt summarizes two tasks comparing GPT-4o mini with GPT-3.5 Turbo: SQL query generation and a portfolio JSON object. The full linked article holds the detailed outputs.
The first task will be to generate syntactically-valid SQL queries. These SQL queries will be executed against the database, and we will examine if the results look correct.
The second task will be to generate syntactically-valid JSON objects. The JSON objects will be used downstream and we’ll be able to determine if the model generated it appropriately.
Key Points Summary
In my side-by-side comparison with GPT-3.5 Turbo, I ran two complex reasoning tasks using my AI-powered algorithmic trading platform, NexusTrade. The results were astonishing:
✅ SQL Queries:
- GPT-3.5 Turbo: Failed to generate a valid query, throwing errors consistently.
- GPT-4o mini: Delivered accurate results on the first try in this reported July 2024 SQL test.
✅ JSON objects:
- GPT-3.5 Turbo: Ran out of context and failed to generate a valid object.
- GPT-4o mini: Successfully generated a valid portfolio and even passed a backtest configuration!
For these two July 2024 tasks, I reported GPT-4o mini succeeding where GPT-3.5 Turbo failed. The 60% cost comparison and 128K context figure are historical claims from this launch, rather than current pricing or a broad task-success rate.
With this advancement, OpenAI is raising the bar, and I believe it will spark fierce competition from other players like Anthropic and Meta. The future of AI is looking brighter and more affordable for all companies leveraging these technologies.
Follow me: LinkedIn | X (Twitter) | TikTok | Instagram
Impressed by Aurora's upgrade? Share this with a friend interested in smarter trading and investing! ♻️
No comments yet.