OpenAI has officially introduced GPT-6, heralding what the company describes as the beginning of the AGI era. During a presentation, OpenAI president Greg Brockman shared insights with reporters, highlighting the model's impressive performance on the ARC-AGI-3 benchmark, where it achieved a score of 98.6%. In contrast, the previous GPT-5.6 Sol barely reached 8%, indicating a significant leap in the model's capacity for handling unfamiliar scenarios and abstract reasoning tasks, now matching human-level performance on the benchmark.
In a groundbreaking advancement, OpenAI promises that employees will no longer need to use a mouse if they choose not to. Previously, developers had to create separate integrations for each application using APIs and connector layers. The new model, Astra, interacts with computers in a human-like manner, autonomously opening browsers, filling spreadsheets, editing documents, and creating 3D models and dashboards without requiring guidance on where to click or what to type.
In the agentic OSWorld 2.0 evaluation, Astra demonstrated a task accuracy of 72.6%, completing each task in an average of 40 minutes. In comparison, GPT-5.6 Sol achieved a 65.7% success rate while taking 75 minutes per task. This means that the "first model of the AGI era" still fails to complete over a quarter of tasks that are typically straightforward for humans.
Pricing for GPT-6 is set to be comparable to Anthropic's Fable line, costing $10 to $50 for every million input/output tokens. However, Brockman noted that with the introduction of GPT-6, the economics of tokens are becoming less relevant; the focus is shifting towards the value of tasks completed by AI in real economic terms.
Additionally, Astra has crossed a critical threshold regarding cyber risk, becoming the first OpenAI model to reach this level. For the initial days following its launch, GPT-6 will be accessible exclusively to corporate clients enrolled in a gated access program. Nevertheless, regular users with a Plus subscription can expect to gain access shortly thereafter.
The question remains: do you think the AGI era has truly begun?
Informational material. 18+.