Tech

The last six months in LLMs: Key updates for businesses

In the past six months, the landscape of large language models (LLMs) has transformed dramatically, marked by rapid advancements and shifting leadership among top models. November 2025 was a pivotal month, often referred to as the 'inflection point,' where coding agents evolved from experimental tools to reliable daily drivers. This period also saw the rise of personal AI assistants, known as 'Claws,' and significant improvements in local models, democratizing AI access. These developments are reshaping how businesses leverage AI technologies.

The November 2025 inflection point

November 2025 marked a critical turning point for LLMs, particularly in coding applications. OpenAI and Anthropic's efforts in Reinforcement Learning from Verifiable Rewards (RLVR) culminated in coding agents that transitioned from 'often-work' to 'mostly-work' status. This shift allowed developers to use these agents as reliable tools for daily tasks without excessive oversight, significantly boosting productivity.

During this period, the title of 'best' model changed hands five times among major players, reflecting intense competition and rapid advancements. Claude Sonnet 4.5, GPT-5.1, and Gemini 3 were among those vying for the top spot, each offering incremental improvements. This dynamic environment underscores the fierce race to maintain AI leadership.

The November inflection point also highlighted the growing capabilities of coding agents, which began to excel in generating high-quality code. These agents became indispensable for developers, handling tasks like code generation and review with minimal errors.

Emergence of personal AI assistants

Personal AI assistants, colloquially known as 'Claws,' have gained attention since late 2025. Originating from the Warelay project, these assistants quickly gained popularity. OpenClaw, a notable example, became a cultural phenomenon, leading to a surge in demand for hardware like Mac Minis to run these systems.

Claws represent a new category of software, offering continuous AI assistance on personal devices. This development has profound implications for businesses, enabling more personalized and efficient workflows. The metaphor of a Mac Mini as an 'aquarium for your Claw' captures the novelty and potential risks of these autonomous agents.

The popularity of Claws highlights the growing trend of integrating AI into everyday business operations. Companies are exploring how these assistants can enhance productivity and streamline processes, making them a valuable asset in the corporate toolkit.

Continue reading

Local models and democratization of AI

The past six months have also seen significant advancements in local models, which are becoming more accessible alternatives to some traditional solutions. Notably, Chinese AI labs released models like GLM-5.1 and Qwen3.6-35B-A3B, which offer impressive performance while running on consumer-grade hardware.

These developments are making AI more accessible by expanding the availability of powerful models to a broader audience. Businesses can now leverage local models to reduce costs, maintain data privacy, and enable offline operations. This shift is particularly beneficial for small and medium enterprises that may not have the resources to invest in high-end AI infrastructure.

As local models continue to improve, they are anticipated to play a significant role in the AI landscape, providing businesses with more flexible and cost-effective solutions for their AI needs.

Challenges and limitations

Despite the advancements, the AI community remains divided on the practical utility of coding agents. While some developers report significant productivity gains, others find these tools unreliable for complex tasks, requiring extensive oversight.

Challenges also persist in the infrastructure required to run advanced models. The high computational costs associated with running powerful agents limit their accessibility to well-funded organizations. This highlights a gap between the democratization of model weights and the actual capabilities available to smaller entities.

Moreover, some benchmarks have become less effective in assessing model capabilities. The AI community continues to seek more relevant and challenging benchmarks to evaluate the true potential of these models.

Future outlook and developments

Looking ahead, the AI landscape is poised for further transformation. The next inflection point may arrive when coding agents can reliably handle complex multi-step tasks across systems. Additionally, as local models approach the capabilities of frontier models, businesses will have more options for integrating AI into their operations.

Reducing the infrastructure costs of running advanced agents will be crucial for broader adoption. As these challenges are addressed, the pace of AI integration into business processes is expected to accelerate, offering new opportunities for innovation and efficiency.

For businesses, the focus will shift from whether to adopt AI technologies to how best to implement them, ensuring they remain competitive in an increasingly AI-driven world.

Frequently Asked Questions

What are the key advancements in LLMs over the past six months?

The last six months have seen significant advancements in LLMs, including the rise of coding agents that are now reliable for daily use, the emergence of personal AI assistants known as 'Claws,' and improvements in local models that make AI more accessible. These developments are reshaping how businesses can leverage AI technologies.

How do personal AI assistants, or 'Claws,' impact businesses?

Personal AI assistants, or 'Claws,' offer continuous AI assistance on personal devices, enhancing productivity and streamlining workflows. They allow businesses to integrate AI into everyday operations, providing a new level of efficiency and personalization. This trend is particularly beneficial for companies looking to optimize their processes.

What challenges do businesses face in adopting advanced AI models?

Businesses face challenges such as the high computational costs of running advanced AI models, which can limit accessibility to well-funded organizations. Additionally, while coding agents have improved, they may still require oversight for complex tasks. Addressing these challenges is crucial for broader AI adoption and integration.