OpenAI has officially announced its cost-efficient small model ‘ChatGPT-4o’ and expects that this launch will expand the range of its applications that are built with AI making the intelligence much more affordable.
ChatGPT-4o scored 82% on MMLU and currently outperforms GPT-4o on chat preferences in LMSYS leaderboards. This model is 60% cheaper than the previous frontier GPT-3.5 Turbo.
Also, it enables a broad range of tasks with low cost and latency and passes a large volume of context to the multiple model calls while with interacting the customers in real-time text responses.
ChatGPT-4o mini supports text and vision in the API by providing support for the text, image, video, and audio inputs and outputs coming in the future.
ChatGPT-4o mini stands out as a highly capable model across multiple domains of artificial intelligence, its performance on various benchmarks indicates superior capabilities compared to other contemporary models. A wide range of applications has stated to provide its performance capabilities:
1. Textual Intelligence And Reasoning Tasks
- Benchmarks: ChatGPT-4o mini scores 82.0%
- Comparison: outperforms Gemini Flash 977.9%0 and Claude haiku (73.8%).
- Implications: it evaluated the model’s ability and understand the reason across different types of textual data,789+540.
2. Mathematical Coding And Proficiency
- Benchmark: MGSM (Mathematical Reasoning )
- Performance: GPT-40 mini scores 87.0%
- Comparison: higher than Gemini Flash (75.5) and Claude Haiku (71.7%).
- Implications: demonstrates superior capability in mathematical reasoning tasks, suggesting its effectiveness in handling complex mathematical queries and calculations.
- Benchmark: HuamanEval (Coding Performance)
- Performance: ChatGPT-4o mini scores 87.2%
- Comparison: Significantly better than Gemini Flash (71.5%) and Claude Haiku (75.9%).
- Implications: highlights its strong coding proficiency, indicating its potential for supporting applications that involve executing and processing code.
3. Multimodal Reasoning
- Benchmark: MU (Multimodal Reasoning Evaluation)
- Performance: ChatGPT-4o mini scores 59.4%
- Comparison: outperforms Gemini Flash (56.1%) and Claude Haiku (50.2%).
- Implications: shows robust performance in tasks that require understanding and reasoning across different types of data modalities, such as combining text with visuals or other sensory inputs.
Built In Safety Measures
ChatGPT-4o mini incorporates built-in safety measures and it ensures maintenance high standard of safety and reliability.
- Pre-Training Filtering: filters out the harmful content to prevent the learning and outputting of harmful information.
- Post-Training Alignment: uses techniques like Reinforcement Learning with Human Feedback to align models behavior.
- Expert Evaluations: helps to evaluate to identify risks by external experts to improve their safety.
- Advanced Safety Techniques: Implement new methods to enhance reliable performance in real-world applications.
- Continuous Monitoring: Open-AI monitors usage to promptly address new risks based on research and feedback.
Availability And Pricing
ChatGPT-4o mini is now available as a text and vision model in the Assistant API, Chat completions API, and Batch API. In ChatGPT, free Plus and Team users will be able to access ChatGPT-4o mini in place of ChatGPT-3.5 and available to enterprise users from next week. GPT-4o mini will available on $0.24 per 1 million tokens which is the lowest amount of the year to subscription.