Author: IBL News

IBL News | New York

Google Gemini’s Gemma open models reached a milestone of 150 million downloads this month.

Omar Sanseviero, a developer relations engineer at Google DeepMind, also revealed that developers have created more than 70,000 variants of Gemma on the AI dev platform Hugging Face.

This collection of lightweight models, built from the same research and technology that powers Gemini 2.0, was launched in February 2024 to compete with another open model like Meta’s Llama. In late April, it exceeded 1.2 billion downloads.

The latest Gemma releases are multimodal, enabling users to work with images and text in 100 languages. Versions are also fine-tuned for particular applications, like drug discovery.

Gemma and Llama have non-standard licensing terms, which some developers say they might be risky for commercial use of the models.

Gemma just passed 150 million downloads and over 70k variants on Hugging Face🚀🚀🚀

What would you like to see in the next Gemma versions?

— Omar Sanseviero (@osanseviero) May 11, 2025

Mikel Amigot, IBL News | New York

OpenAI launched a paid software engineering agent called Codex yesterday. The agent iteratively runs tests on its code until it achieves passing results.

Vibe coding tools have surged in popularity in recent months, and OpenAI is attempting to capture a share of that market segment. In early May, the maker of ChatGPT reportedly closed a deal to acquire Windsurf, the developer behind another popular AI coding platform, for $3 billion. The launch of Codex shows that OpenAI is also building its own AI coding tools.

Issued as a research preview, Codex can work on many tasks in parallel, write features, answer questions about the user’s codebase, fix bugs, and propose pull requests for review.

Each task runs on a sandboxed virtual computer in the cloud, preloaded with the repository.

According to OpenAI, it generates code that closely mirrors human style and PR preferences, adheres precisely to instructions, and can iteratively run tests until it receives a passing result.

Codex, accessible through the sidebar in ChatGPT, is powered by Codex-1, a version of OpenAI o3 AI reasoning model optimized for software engineering tasks.

OpenAI is also updating Codex CLI, the company’s recently launched open-source coding agent that runs in the user’s terminal, with a version of its o4-mini model that’s optimized for software engineering.

In OpenAI’s API, it costs $1.50 per 1M input tokens (750,000 words) and $6 per 1M output tokens.