Deep Learning Weekly: Issue 427

This week in deep learning, we bring you Introducing Claude Haiku 4.5, The State of Open Models, and a paper on TOUCAN: Synthesizing 1.5M Tool-Agentic Data from Real-World MCP Environments.

You may also enjoy How a Gemma model helped discover a new potential cancer therapy pathway, A Definition of AGI, a paper on Training Software Engineering Agents and Verifiers with SWE-Gym, and more!

As always, happy reading and hacking. If you have something you think should be in next week’s issue, find us on Twitter: @dl_weekly.

Until next week!

Industry

Introducing Claude Haiku 4.5 \ Anthropic

Anthropic introduced Claude Haiku 4.5, the latest version of their smallest model with state-of-the-art computer-use and coding capabilities at one-third the cost.

How a Gemma model helped discover a new potential cancer therapy pathway

Google launched a new 27 billion parameter foundation model for single-cell analysis built on the Gemma family of open models.

Method teaches generative AI models to locate personalized objects

Researchers from MIT and elsewhere have introduced a new training method that teaches vision-language models to localize personalized objects in a scene.

Introducing Veo 3.1 and advanced capabilities in Flow

Google introduced a significantly updated version of their filmmaking tool called Veo 3.1, which brings richer audio, more narrative control, and enhanced realism.

Announcing Tinker - Thinking Machines Lab

Thinking Machines Lab launched Tinker, a flexible API for fine-tuning language models, which provides researchers control over the algorithms and data without having to think about the complexity of distributed training.

MLOps & LLMOps

Legacy data to RAG : Modernise Your Apps with Amazon Sagemaker Unified Studio

A blog post about modernizing applications by integrating Amazon SageMaker Unified Studio and Weaviate to facilitate efficient hybrid search and context-aware RAG queries on legacy data.

Learning

The State of Open Models - by Nathan Lambert

An insightful talk reflecting on the landscape of open models, discussing the dominance of Chinese AI, the fade of Llama, and strategies for steering the future direction of open-source AI.

A Definition of AGI

A quantifiable definition of Artificial General Intelligence (AGI) as matching a well-educated adult’s cognitive versatility, operationalized through the Cattell-Horn-Carroll theory and psychometric batteries.

Evaluating Gemini 2.5 Deep Think’s math capabilities

A critical article evaluating Gemini 2.5 Deep Think’s math capabilities, noting its impressive skill in complex computations and background knowledge, but highlighting its limitations in creative problem-solving.

Libraries ...