Key Concepts
- GPT-4o Mini and GPT-3: New models from OpenAI, with a focus on tool calling and multimodal capabilities.
- Tool Calling: The ability of the models to use external tools like image generation and Python interpretation.
- Multimodal: The ability of the models to process and reason with different types of data, including text and images.
- ADER Polyglot Benchmark: A benchmark used to evaluate the performance of language models.
- Codeex CLI Tool: A new open-source terminal interface from OpenAI for coding, similar to Claude Code or Aider.
- Ninja Chat: An all-in-one AI platform offering access to multiple AI models.
OpenAI's New Models: GPT-4o Mini and GPT-3
OpenAI has launched their GPT-4o Mini and GPT-3 models. The GPT-3 model is the same model that was announced in December but wasn't made available until now. Both models are specifically trained for tool calling, allowing them to use tools like image generation and Python interpretation within chat GPT.
Performance Benchmarks
- GPT-3: In the Ader Polyglot benchmark, GPT-3 scores 81.3%, which is about 10% higher than Gemini 1.5 Pro.
- GPT-4o Mini: Scores 58.2% in the same benchmark, which is 15% lower than Gemini 1.5 Pro. It also doesn't beat Claude 3 Sonnet.
- The high variant of GPT-4o Mini may cost the same as Gemini 1.5 Pro but offers worse performance.
- GPT-3 is noted to be very expensive.
Thinking with Images
One notable feature is the ability of these models to "think with images," integrating images directly into their chain of thought, including zooming into images.
Pricing
- GPT-4o Mini: Costs $0.003 per 1K tokens for input and $0.015 per 1K tokens for output.
- GPT-3: Costs $0.04 per 1K tokens for input and $0.20 per 1K tokens for output.
Ninja Chat: An Alternative AI Platform
Ninja Chat is presented as an all-in-one AI platform that provides access to top AI models like GPT-4o, Claude 3 Sonnet, and Gemini 1.5 Flash for $11 per month. It features an AI playground for comparing responses from different models and a mind map generator. The basic plan includes 1,000 messages, 30 images, and five videos monthly. A discount code "king25" for 25% off any plan or "king40yearly" for 40% off annual subscriptions is provided.
Codeex CLI Tool
OpenAI has also launched a new open-source Codeex CLI tool, similar to Claude Code or Aider. It's a terminal interface that allows coding, making changes to files, and is compatible with Mac and Linux. It uses the OpenAI API and is built with TypeScript.
Key features:
- Open Source: Allows for community contributions and modifications.
- Chat-Driven Development: Enables developers to use chat GPT-level reasoning within their terminal.
- Integration: Works with version control systems.
- Auto Mode: Automates certain coding tasks.
The tool is named after OpenAI's first coding model, OpenAI Codeex.
Conclusion
The GPT-4o Mini model might cost the same as Gemini 1.5 Pro but provide worse results based on benchmarks. The GPT-3 model is good but expensive. The Codeex CLI tool is a welcome addition, especially due to its open-source nature. The speaker plans to test these models further and provide updates in future videos. Free ways to access these models may become available through platforms like GitHub models, requesty, and Windsurf.
AI summaries can miss context or contain errors. Check important details against the original video.
