OpenAI Bets on Instant Response

ChatGPT has a new default model. OpenAI began late this week fully deploying GPT 5.5 Instant, which replaces the previous version for regular users. The company promises that the model is "smarter, more accurate, and more personalized." The most significant change, however, is speed — GPT 5.5 Instant has a significantly shorter response time than its predecessors, meaning smoother conversations without long waiting periods.

The model is available for free in the free tier, albeit with ads that should start appearing in the ChatGPT interface. OpenAI's goal is clear: to maintain and expand the base of 900 million weekly users, which according to available data has not moved much in the last three months. The fast and free version is intended to serve as a gateway to paid subscriptions like ChatGPT Plus, Team, or Enterprise.

From a technical perspective, GPT 5.5 Instant utilizes extensive computing infrastructure, into which OpenAI invests tens of billions of dollars annually. Lower latency is crucial, especially for agent functions — i.e., scenarios where AI itself performs multi-step tasks such as document searching, planning, or basic coding. The faster the model responds, the more naturally it acts as a true digital assistant.

For Czech users, it is important that ChatGPT, including GPT 5.5 Instant, is fully available in the Czech Republic and supports Czech at a very good level. The free tier is sufficient for common queries, email writing, or translations, while the paid version for approximately 20–25 USD per month unlocks higher limits and access to tools.

Google to Prepare Fast Gemini 3.2 Flash

While OpenAI is already deploying, Google is preparing a counterattack. At the conference Google I/O 2026, which will take place on May 19, the introduction of Gemini 3.2 Flash — a lightweight and fast variant of the flagship model — is expected. Along with it, Gemma 4, a new version of the open-source model designed for developers, should also arrive.

Google builds speed on its own TPU infrastructure (Tensor Processing Units), which gives it the advantage of vertical integration — from chips to cloud to applications. This allows Gemini to deliver responses in milliseconds to billions of users across the Google ecosystem: Search, Gmail, Docs, Drive, YouTube, Maps and other services.

A key difference compared to OpenAI is deep integration into existing applications. While ChatGPT requires the user to open a separate application or website, Gemini works directly in the environment people already use daily. For Czech users, this means they will automatically get faster AI — without needing to install a new tool. Google also continues its collaboration with Apple, which later this year could mean better integration of Gemini with Siri.

For developers, the news about Gemma 4 is important. Google's open-source models allow AI to be run locally or in one's own cloud, which is an advantage for European companies in terms of GDPR and data sovereignty.

Anthropic Introduces Orbit and Addresses Computing Capacity

The third player this week is Anthropic. At the conference Code with Claude it introduced Orbit — a proactive AI assistant designed for the service Claude Cowork. Orbit differs from classic chatbots in that it doesn't sit passively and wait for a query. Instead, it analyzes context, anticipates user needs, and offers solutions before being asked.

This approach has great potential for regular users, not just for programmers. Anthropic is growing rapidly — the company is reportedly aiming for 30 billion USD in annual revenue (ARR) and in some metrics even surpasses OpenAI. Furthermore, the Claude model is at the core of other companies' services, such as Perplexity, which means its impact extends far beyond Anthropic's direct customers.

However, growing demand brings a problem. Anthropic cannot keep up with expanding its computing capacity, and therefore this week announced an agreement with SpaceX/xAI to use the Colossus supercomputer. This "alliance of necessity" will allow Anthropic to absorb rapid growth, but at the same time indicates how critical access to computing resources is in the current AI era.

For Czech users, Claude is available via a web interface or API, with Czech fully supported. Claude excels particularly in long contexts and response safety.

What This Means for Regular Users and Businesses

The race for speed is not just a marketing trick. Shorter response time directly affects the usefulness of AI in two key scenarios:

• Everyday productivity: When writing emails, editing documents, or doing quick translations, waiting for a response disrupts the workflow. A fast model acts more like an extension of thought than a standalone tool.
• Agent workflow: When AI performs a multi-step task — for example, booking a flight, writing code, and testing it — every second of delay multiplies. Instant models enable smoother human-machine collaboration.

For Czech companies, this development is good news. Competition among OpenAI, Google, and Anthropic drives prices down and increases the quality of free tiers. At the same time, Czech language support is rapidly improving — although it is still true that English queries often lead to more precise results for the most complex tasks.

Where to Try the New Models

GPT 5.5 Instant is already available to most ChatGPT users for free at chat.openai.com or in the mobile app. The paid version of ChatGPT Plus costs approximately 20–25 USD per month.

Gemini 3.2 Flash will likely be introduced on May 19 at Google I/O. The current version of Gemini can be tested at gemini.google.com, with Google One AI Premium costing around 20 USD per month.

Anthropic Claude and Orbit are available at claude.ai. To access the latest features, a subscription to Claude Pro for approximately 20 USD per month is required.

All three services are legally available in the Czech Republic and are subject to the EU AI Act, which means guarantees of transparency and user protection. However, for companies processing sensitive data, it is still more appropriate to consider local deployment of open-source models, such as Gemma 4 will be, or the use of European cloud providers.