O3, O4 Mini & Codex CLI + Free APIs The MAGIC is MISSING from these new LAUNCHES?!
AI Summary
Summary of OpenAI’s New Mini Models
- Model Overview
- OpenAI launched two mini models: 03 and 04.
- 03 model is an improved version of the December model, now available.
- Both models are multimodal and capable of tool calling.
- Performance
- 03 Model: Scores 81.3% on Ader’s Polyglot benchmark, outperforming Gemini 2.5 Pro by approximately 10%.
- 04 Mini Model: Scores 58.2%, underperforming Gemini 2.5 Pro by about 15%.
- Performance evaluations suggest 04 Mini may not be worth its cost relative to competitors.
- Model Capabilities
- Both can integrate and reason with images: can “think with images” and perform tasks such as zooming in on them.
- Pricing
- 04 Mini: Costs about 0.275 for input with caching, and $4.40 for output.
- 03: More expensive at 2.50 with caching, and $40 for output.
- CLI Tool - Codeex
- New open-source CLI tool for coding tasks, allowing file manipulation and execution of code.
- Compatible with Mac and Linux only, utilizing Apple seatbelt for security.
- Supports the OpenAI API and provides a terminal interface optimized for developers who use terminal for coding.
- Can be installed via npm.
- General Observations
- Model performance is decent but not extraordinary; requires further testing to validate claims.
- Free access options include GitHub and Windsurf.
- Viewers encouraged to share experiences with the new models for better insights on performance.