The Batch Issue 323, published on October 15, 2025, presents several major advances in artificial intelligence, with a particular focus on the optimization of infrastructures and ML development processes.

Inference cost reduction by DeepSeek

DeepSeek announces significant innovations in reducing the inference costs of AI models. This advance is particularly important in a context where the operational costs of large language models constitute a major obstacle to their large-scale deployment. The techniques developed by DeepSeek aim to optimize the use of computational resources during the inference phase, thereby making AI applications more economically accessible. This innovation is part of a broader industry trend aimed at democratizing access to AI technologies by reducing financial barriers.

Strategic OpenAI-AMD partnership

OpenAI strengthens its ties with AMD, marking a significant evolution in the AI hardware landscape. This partnership reflects a desire for diversification beyond the traditional NVIDIA ecosystem and could have significant implications for the industry. The collaboration likely aims to develop hardware solutions specifically optimized for OpenAI's training and inference workloads. This strategic alliance could also help ease tensions related to limited GPU availability and foster healthier competition in the AI hardware market.

Fine-tuning simplification by Thinking Machines

Thinking Machines presents tools aimed at simplifying the fine-tuning process for machine learning models. Fine-tuning remains a crucial but often complex step in adapting pre-trained models to specific use cases. The proposed solutions seek to make this process more accessible to technical teams without deep ML expertise, while maintaining the quality of results. This simplification is essential to accelerate AI adoption in enterprises and enable more organizations to benefit from customized models.

Context and implications

This issue of The Batch reflects current trends in the AI industry: cost optimization, hardware infrastructure diversification, and democratization of development tools. These three axes are complementary and essential for moving from experimental AI to AI truly deployed at scale. The emphasis on operational efficiency and accessibility suggests a maturation of the sector, with growing focus on the economic viability and practicality of AI solutions.

The newsletter also covers other topics ("and more..."), reflecting the richness and diversity of innovations in the field of artificial intelligence during this period of October 2025.