Last month in AI – April 2025
AI-driven Newsletter
Knock, knock! It’s a second edition of Last month in AI for 2025!
This month has been packed with new releases and big updates – from Meta’s ambitious Llama 4 (and the mixed reactions it sparked) to Google’s fascinating DolphinGemma project, plus fresh developments from OpenAI and Qwen. Whether you’re here to track the latest models, discover new tools, or just stay in the loop with what’s shaking up the AI world, we’ve got you covered.
Let’s dive right in and explore the highlights from last month!
Models
Llama 4
Meta released a new version of their Llama model with two different sizes: Scout, Maverick, and teased the largest Behemoth, which is still under training.
Unfortunately, the first reviews are not so good for Llama 4. LmArena confirmed that their Llama 4-Maverick was a version “customized for human preference”. The community is not happy about the performance of the model, and even the Meta employees are distancing themselves from the work of the team responsible for Llama 4.
Qwen3
New batch of Qwen models. This time, they are focused on reasoning. They also included 2 MoE models: Qwen3-235B-A22B and Qwen3-30B-A3B. Based on benchmarks, they look promising, and from my short tests, they are performing well.
ChatGPT 4.1 variants
New models launched exclusively in API. Names of new models are GPT-4.1, GPT-4.1-mini and GPT-4.1-nano. Apparently 4.1 is supposed to be better than 4.5. Good job with the naming OpenAI!
ChatGPT o3 and o4-mini
Finally, o3 has arrived! With a smaller brother o4-mini. GPT-o3 is supposed to be a new SOTA model that excels in reasoning tasks(it’s supposed to think longer than it’s predecessors). It should handle better scientific and programming tasks. GPT-o4-mini on the other hand is going hard when using tools and when given access to the python interpreter it gets near perfect scors in the major scientific benchmarks.
DolphinGemma
LLM understanding dolphin language? Why not! Google with Wild Dolphin Project and Georgia Tech announced their collaboration on DolphinGemma model which is trained to understand dolphins language.
Gemini 2.5 Flash Preview
Gemini 2.5 Flash is Google’s latest cost-effective, thinking model designed for fast performance on complex tasks. It’s a hybrid reasoning model, meaning developers can choose to turn thinking on or off, and can also set a “thinking budget” to control how much processing time is used for reasoning.
Gemma-3 QAT
Gemma 3 QAT models are versions of Google’s Gemma 3 AI optimized using Quantization-Aware Training. This technique drastically reduces the models’ memory (VRAM) requirements while preserving high-performance quality, unlike standard quantization methods. Consequently, even large models like Gemma 3 27B can run effectively on accessible consumer hardware, such as an NVIDIA RTX 3090 GPU.
Hardware
This month didn’t have any major hardware releases. We’re still waiting for the Quadro Pro RTX Blackwell series, even though they were announced in the middle of March. The latest news is that it will be getting them early in May.
NVIDIA GeForce RTX 5060 Ti (Blackwell GPU)
NVIDIA released the next GPU from Blackwell family. This time with either 8GB or 16GB of memory. I think it’s a shame that second highest GPU, the 5080, has the same amount of memory.
Other
OpenAI Codex CLI
OpenAI Codex CLI is a recently released, open-source command-line tool that functions as an AI coding assistant directly within your terminal. It leverages current OpenAI models (like o4-mini) to understand natural language prompts, helping you read, modify, and execute code locally with configurable safety modes.
xAI Grok Studio
Elon Musk’s new AI company xAI gave its chatbot Grok a creative boost by launching Grok Studio – a drag-and-drop canvas for making custom AI workflows. This intuitive interface lets you visually chain together prompts, documents, and data sources to build everything from interactive reports to mini-apps.
Summary
That wraps up this month’s round-up of AI’s most significant moves and moments! We hope you found insights that sparked your curiosity or inspired new ideas.
Stay tuned for next month’s edition, the pace of innovation isn’t slowing down, and we’ll be here to help you keep up.
Until next time – keep exploring, keep learning, and check out our other articles!
Also, don’t forget to join Scalac’s Talent Pool!
Check more here https://scalac.io/blog/scala-rust-devops-frontend-careers/
Authors
An AI expert and Scala developer at Scalac, providing ongoing analysis of key developments in artificial intelligence. Scalac's go-to specialist for AI trends and applications. His work bridges the gap between AI research and practical business implementation, making him a trusted voice not only among all the blog posts here, but in the AI community in general. Also, a proud owner of a Czechoslovakian Wolfdog, one of the closest-to-wolf dog breeds that you can legally own.
