OpenAI has introduced GPT-4o mini, a smaller and more cost-effective version of its powerful GPT-4o model.
GPT-4o mini is touted as “essentially the most cost-effective small model available on the market” and is significantly cheaper than the competition.
Developers pay just $0.15 per million input tokens and $0.60 per million output tokens, in comparison with $5.00 and $15.00, respectively, for GPT-4o.
Olivier Godement, OpenAI’s Head of Product, API, discussed the potential of the model with VentureBeat: “The cost per intelligence is so good that I expect it for use for every kind of customer support, software development, creative writing, and every kind of tasks.”
Despite the “mini” icon, GPT-4o mini offers impressive capabilities. It outperforms GPT-3.5 Turbo in various benchmarks and might handle each text and visual input.
OpenAI reports that GPT-4o mini achieves a rating of 82.0% on the Massive Multitask Language Understanding (MMLU) benchmark, beating competitors equivalent to Google's Gemini 1.5 Flash (77.9%) and Anthropic's Claude 3 Haiku (73.8%).
The model is meant to interchange GPT-3.5 Turbo for ChatGPT Plus and Teams subscribers, offering users a more powerful model at no additional cost.
Early adopters, including startups Ramp and Superhuman, have reported promising results for tasks equivalent to receipt categorization and personalized email responses.
OpenAI wants to verify the safety of GPT-4o mini
While OpenAI pushes the boundaries with the capabilities and affordability of GPT-4o mini, it doesn’t skimp on security. It uses the identical mechanisms it developed for the larger GPT-4o model.
OpenAI also brought in over 70 experts from fields equivalent to social psychology and disinformation to place GPT-4o through its paces.
These specialists helped discover potential risks so the team could address issues before they became problems. The findings were incorporated into GPT-4o mini.
OpenAI also introduced what it calls the “command hierarchy” method, which “helps improve the model's resilience to jailbreaks, prompt injections, and system prompt extractions. This makes the model's responses more reliable and safer to make use of in large applications.”
This might be a marketing slogan for corporate users who need to avoid erroneous results and hallucinations in any respect costs.
In the longer term, OpenAI plans to expand the capabilities of GPT-4o mini, including the flexibility to generate image, audio and video output. The model can even be available via Apple Intelligence this fall, coinciding with the discharge of iOS 18.
While GPT-4o mini is kind of exciting, OpenAI has suffered setbacks in other areas. The company recently delayed the discharge of the speech and emotion reading features for ChatGPT, citing the necessity for extra security testing.
People were amazed when the corporate demonstrated GPT-4o and its speech synthesis, but things have calmed down since then.
Still, GPT-4o mini proves that the parents at OpenAI are still working hard despite some recent controversies.
