2026
GPT-5
OpenAI launches GPT-5: the first model that consistently reaches human level across a broad spectrum of cognitive tasks, with strong reasoning and multimodal capabilities.
The next frontier
In 2026, OpenAI launched GPT-5 — a model that represented a significant step beyond GPT-4o in both capability and reliability. On standard benchmarks covering reasoning, coding, mathematics, science, and language understanding, GPT-5 consistently performed at or above the level of a highly educated human professional. It combined text, image, audio, and video understanding in a single unified architecture.
What changed
GPT-5's most notable improvement over its predecessors was consistency and reliability. Earlier models were capable of impressive performance on specific tasks but prone to unexpected failures, hallucinations, and inconsistencies. GPT-5 showed markedly reduced hallucination rates, better calibration of uncertainty, and more reliable adherence to instructions across long, complex tasks. These reliability improvements were at least as important as the raw capability gains for enterprise adoption.
The AGI debate
GPT-5's release reignited debate about artificial general intelligence. OpenAI stopped short of claiming AGI, but noted that GPT-5 met several of its internal capability thresholds for what it called "broadly safe AGI." Critics argued that benchmark performance was not equivalent to general intelligence, and that GPT-5 still lacked genuine understanding, common-sense reasoning in novel physical situations, and reliable long-horizon planning. The debate reflected genuine uncertainty about what AGI means and whether it had been achieved.