GPT-5 arrives with native multimodal capacity and persistent memory
OpenAI launched GPT-5, the company's first model with native unified vision, audio and text, as well as long-term memory between conversations.
OpenAI today has revealed GPT-5, the company's most ambitious model to date. The launch brings two fundamental changes: native multimodality and persistent memory.
Real multimodality
Unlike the GPT-4o, where the modalities were processed by separate modules, the GPT-5 was trained from the beginning with text, image, audio and video as unified inputs. In practice, this means that the model understands visual and auditory context with the same depth as it understands text.
In demonstrations, OpenAI showed the model analyzing a 30-minute video of a meeting, identifying each participant by voice, transcribing speeches and summarizing decisions — all in real time.
Persistent memory
The GPT-5 introduces a memory system that persists between conversations. The model recalls preferences, ongoing projects and professional context of the user. OpenAI ensures that this memory is encrypted end-to-end and can be erased at any time.
Pricing and access
GPT-5 is available today for Plus subscribers ($20/month) with a limit of 40 messages per day. Access via API starts at $5/M tokens input and $15/M tokens output.
Sam Altman stated on the X: "GPT-5 is the model I wanted to launch two years ago. Now, finally, we can."
Source: openai.com
Written by
Lucas MontarroiosSou Lucas Montarroios e dediquei os últimos 15 anos à linha de frente de operações de telecom e data centers. Minha carreira sempre foi pautada por um foco implacável: transformar tecnologia e cenários críticos em oportunidades reais de negócio. No novidades.ia.br, trago essa visão executiva para o universo da IA. Especialista em produtos, mercado e ferramentas práticas de IA. Minha missão aqui é filtrar o ruído do mercado, analisando benchmarks, estratégias de grandes empresas e ferramentas práticas.
Related articles
Mistral Large 4 (Le Chonk) sai com pesos abertos e estreia na Microsoft Foundry
A Mistral lançou o Mistral Large 4, apelidado de Le Chonk, como modelo de pesos abertos, e a Microsoft é a primeira nuvem a oferecê-lo via Microsoft Foundry.
OpenAI GPT no Amazon Bedrock: inferência global chega a cinco países da Ásia-Pacífico
A AWS passou a oferecer os modelos GPT da OpenAI pelo Amazon Bedrock, com inferência cross-region em Tailândia, Malásia, Singapura, Indonésia e Taiwan.
Meta lança Muse Wonder e Muse Wonder Full e volta à disputa de fronteira em IA
A Meta anunciou os modelos de fronteira Muse Wonder e Muse Wonder Full, treinados no data center Prometheus, em Ohio, com os recursos Active Memory e Active Focus voltados a agente...