The world of artificial intelligence is currently at a turning point. OpenAI, a company long known for its closed approach to model development, took a step no one expected: it released its first open models under the Apache 2.0 license. These models, designated as GPT-OSS, open the door to running AI locally without cloud dependency, while simultaneously paving the way for a new era of ChatGPT that will no longer be just about text responses.

The end of the chat-only era: Why is “chat dead”?

According to information from Wired magazine, the opinion circulating within OpenAI’s leadership itself is that “chat is dead.” This quote reflects a deeper paradigm shift. The traditional interaction where a user asks a question and waits for an answer is being replaced by the concept of AI agents.

OpenAI’s goal is to create a superapp. This means that ChatGPT will become a central brain that will not just respond, but actively execute tasks. Thanks to the integration of the Codex tool for programming and the ability to leverage autonomous agents, the system will be capable of planning complex workflows – from writing code to managing your digital tasks in real time. This shift is driven by the effort to increase profitability ahead of the company’s planned initial public offering (IPO), where OpenAI wants to focus primarily on the lucrative enterprise clientele.

GPT-OSS: Cutting-edge AI right in your laptop

While ChatGPT is heading to the cloud as a complex system, the GPT-OSS series is intended for those who want control over their data and computing power. OpenAI released two models with different parameters:

• GPT-OSS-120B: A massive model that achieves performance comparable to the o4-mini model. Efficient operation requires a professional workstation with a GPU of at least 80 GB VRAM (e.g., NVIDIA H100).
• GPT-OSS-20B: A more compact variant offering performance on par with the o3-mini model. The key advantage is that this model can be run even on a standard laptop with 16 GB RAM.

These models use the Mixture-of-Experts (MoE) architecture, meaning that only a portion of the parameters is activated for each query, saving computational power and increasing speed. Moreover, both models support chain-of-thought reasoning, allowing users and developers alike to follow the logical path the model took to reach its result – which is crucial for transparency and debugging.

Performance comparison: GPT-OSS vs. the competition

With this move, OpenAI has directly confronted the leaders of the open market, such as Meta with its Llama models or China’s DeepSeek. While Llama excels in general use, GPT-OSS appears to be more optimized for specific tasks such as programming and tool use. In logical reasoning benchmarks, the 20B version shows surprisingly low latency, making it an ideal candidate for Edge AI applications.

NVIDIA and optimization for consumer hardware

To ensure that local deployment isn’t merely theoretical, OpenAI partnered with NVIDIA. The new models are optimized for graphics cards from the GeForce RTX and RTX Pro families. Thanks to support for the new MXFP4 format, the GPT-OSS-20B model can generate up to 250 tokens per second on an NVIDIA RTX 5090 – a speed perceived as instantaneous.

Practical impact: What does this mean for Czech users and businesses?

This development has several major implications that will be felt in the Czech Republic as well:

1. Privacy and GDPR: For Czech companies handling sensitive data (banking, healthcare, public administration), the ability to run the GPT-OSS model locally on their own server or workstation is absolutely critical. Data never leaves the corporate network, making it easier to comply with EU regulations and personal data protection.
2. Cost and accessibility: While you will pay for ChatGPT Plus (expected subscription around $20/month), the GPT-OSS models are free to download under the Apache 2.0 license. This opens the door for Czech startups and developers to build their own applications without high API call costs.
3. Language support: OpenAI has a long-standing tradition of quality localization. It is highly likely that both the ChatGPT superapp and the GPT-OSS models will have excellent Czech language support, which is essential for the Czech market.

For the average user, this means that the line between “searching for information” and “executing tasks” is blurring. AI will no longer be just a text box, but a digital assistant that helps you plan a trip, write code for your website, or automate routine emails.