OpenAI’s GPT-5 model is nearing release, with early testers praising its coding and problem-solving skills, though experts say the leap from GPT-4 may be smaller than previous upgrades.

Two early testers said the model performs well in coding and technical problem-solving but shows more modest gains over GPT-4. They spoke under non-disclosure agreements.

OpenAI has not confirmed a release date, but industry watchers expect the launch soon.

GPT-5 follows GPT-4’s 2023 debut, which set new benchmarks in reasoning and exam performance. By contrast, GPT-5’s training faced data shortages and risks of hardware failures, both common in large-scale AI development.

To improve output, OpenAI is using “test-time compute”, a method of boosting processing power during execution. OpenAI CEO Sam Altman has said future models will combine this with traditional training.

Boris Power, OpenAI’s head of Applied Research, wrote on X, “Excited to see how the public receives GPT-5.”

Navin Chaddha, managing partner at venture capital firm Mayfield said, “The hope is that GPT-5 will unlock AI applications that move beyond chat into fully autonomous task execution.”

The launch comes amid rising competition. Google, Anthropic and Meta have all released models rivaling GPT-4’s performance.

Three years after ChatGPT’s debut, OpenAI is under pressure to deliver its next leap forward.

Copied