OpenAI is preparing to launch GPT-5 as early as August 2025, marking the next major advancement in artificial intelligence development. The new model is expected to unify reasoning and multimodal capabilities, enhancing overall power and coding performance. Evidence of GPT-5's imminent release has been found in the cache data of the ChatGPT macOS app, including components named gpt-5-auto and gpt-5-reasoning, which likely serve as the model router and reasoning engine. Despite the anticipation, sources indicate that GPT-5's performance improvements will be more incremental compared to the substantial leaps seen between earlier models such as GPT-3 in 2020 and GPT-4 in 2023. OpenAI researchers have faced technical challenges in 2025 but have employed techniques like reinforcement learning and universal verifiers to mitigate slowing gains in model performance. The AI community and major tech companies including Microsoft, Meta, Google, Amazon, Alibaba, and Nvidia are closely watching the rollout of GPT-5.
GPT-5 is coming and it'll be good. But, OpenAI and others have been facing slowing gains, which they've been able to find workarounds to with reinforcement learning, "universal verifiers" and other techniques. More here w/ @erinkwoo @amir: https://t.co/KGDdsH9ufu
Source: GPT-5 improvements won't be comparable to the leaps in performance of earlier models, such as between GPT-3 in 2020 and GPT-4 in 2023 (The Information) https://t.co/dqZirHvKTG https://t.co/S9uEe2qvtU https://t.co/ZOzeer1FAj
GPT-5 is good. But model performance gains are still slower than in past years and this year has been a technically challenging one for OpenAI researchers. The inside story here... https://t.co/uWzL7GofRF