OpenAI has officially launched GPT-5, the most powerful model to date, and is providing it for free to all users.
After months of waiting, OpenAI has officially introduced GPT-5, the latest generation of large language models in the context of fierce competition with Meta, Anthropic, and Google. CEO Sam Altman describes this as an "important upgrade compared to GPT-4 and a significant step towards AGI", while emphasizing the model's ability to "write a software program from scratch".
In a practical example, GPT-5 wrote over 400 lines of code within two minutes from a request simulating the Bernoulli effect, demonstrating impressive programming capabilities. Altman compared: "GPT-4 was like talking to a college student. Now, GPT-5 is like talking to an expert."
Context Window Challenges
Although GPT-5 has a 400,000 token context window, four times that of GPT-4, this number still falls short of competitors like Claude 3 and Gemini 1.5 with the ability to process over 1 million tokens. This is a crucial factor determining the amount of information the model can "memorize" and process in a single interaction, including long documents, complex codebases, and extended conversations.
This disparity makes GPT-5 seem to be "catching up" rather than "overtaking" its competitors. Claude can currently receive an entire book series or massive code repository and still respond coherently according to context.
GPT-5 brings a series of new features, including video capabilities that allow the model to "observe" users via camera and integration with Gmail, Google Calendar, to be launched next week. The new model is available to all users from today, with free users being limited in volume and switching to a "mini" version when reaching a threshold.
OpenAI particularly focuses on enterprise applications with enhanced reasoning capabilities through a "Unified Reasoning Architecture". The company eliminates the trade-off between speed and depth of analysis, allowing users to obtain quick answers without losing accuracy.
In terms of performance, OpenAI affirms that GPT-5 is "faster, more reliable, and more accurate" than the previous version, and is the "best programming model to date". Notably, the hallucination rate has significantly decreased, addressing a major challenge in LLM deployment for enterprises.
The important breakthrough lies in a new synthetic data training method. "Our breakthrough is not about creating more data, but creating the right type of data," OpenAI explains, emphasizing superior effectiveness compared to traditional data collection methods.
GPT-5 also integrates a personalized memory system, allowing chatbot behavior to be adjusted according to specific user requirements across multiple conversations. In programming, the model achieves the highest performance on SWE-bench and supports three different variants to optimize token costs.
With 700 million weekly active users of ChatGPT, compared to Google Gemini's 47 million, OpenAI maintains its leading position. However, Meta AI also reaches 700 million users through integration into Instagram and WhatsApp, creating a fierce competition.
OpenAI is currently valued at $300 billion after raising $57 billion through 11 funding rounds, with the potential to increase to $500 billion if a new equity sale is executed. The company expects to reach $20 billion in revenue by the end of next year after doubling in the first 7 months.
Altman first confirmed that OpenAI began developing GPT-5 in 2023, with the actual training phase starting a year later. Previously, the company had released two open-weight models – gpt-oss-120b and gpt-oss-20b – under an open license for developers wanting to run models locally.
