by Tan Aik Keong (AK)
Late last year, China's DeepSeek became a global talking point on the strength of its architectural innovation and low-cost training, earning comparisons to a "Chinese Sputnik moment." Then, in August 2025, OpenAI formally launched ChatGPT-5, loudly billing it as "the smartest model in the world" and rolling it out broadly to users. This isn't just a contest over which model is technically stronger — it's a clash between two AI ecosystem strategies, and between the international influence each side is chasing.
Multimodal reach and raw benchmarks
ChatGPT-5 unifies text, image, audio and short-video input in a single model, with an ultra-long context window (up to 200K tokens), real-time voice conversation and tool-use agents — positioning it as an "everything assistant." DeepSeek, by contrast, remains primarily text-focused; it has open-sourced vision modules, but hasn't yet unified them into as seamless a multimodal experience. On general language understanding and reasoning, ChatGPT-5 posts standout benchmark scores — 94.5% on MMLU, for instance — with a clear lead in maths, logic and coding. DeepSeek's models sit closer to GPT-4-class performance, though its R1 model's pure reinforcement-learning approach produces a striking "self-reflective" style of reasoning in its own right. On balance, ChatGPT-5 holds a general-intelligence edge, but DeepSeek isn't outclassed on complex reasoning tasks.
Chinese-language fluency and strategy
DeepSeek's clear advantage is in Chinese-language and local fluency — many Chinese users find it noticeably more natural on classical poetry, policy text and similar material, and its open-source model means it can be tuned without restriction. ChatGPT-5's training data spans many languages, but tighter content controls on Chinese-language output make it feel less natural than DeepSeek's in some contexts. Strategically, DeepSeek has gone fully open source — weights and code both public — fuelling an active developer community that's pushed daily active users past 30 million. ChatGPT-5 stays closed source but is available to hundreds of millions of users, and has moved quickly to embed itself into Microsoft, GitHub and other enterprise ecosystems, closing the loop on monetisation. Open-source scale versus closed-source polish: each has its own edge.
What it means for the US-China AI race
From the US side, ChatGPT-5's launch is OpenAI's forceful answer to China's AI rise — not just to defend its technical lead, but to compete for ecosystem resources and market influence. From China's side, DeepSeek's success has boosted confidence in, and the international standing of, homegrown AI, spurring other players — Baidu, Alibaba, ShanghaiTech and others — to accelerate their own multimodal efforts. DeepSeek may well fold in vision and further reinforcement-learning work (an R2 model, perhaps) to carve out its own differentiated path forward. Zoom out, and this contest isn't just about who has the stronger model — it's about who gets to set the standards and rules the rest of the AI ecosystem runs on. Some countries will likely gravitate to the strongest closed-source model; others will lean into the open-source camp, and together that pushes the AI world toward a more multipolar shape.
Closing
ChatGPT-5 is, without question, the current high-water mark of AI technology — but DeepSeek's strength in Chinese-language understanding, open-source ecosystem building and reasoning innovation shouldn't be underestimated either. Each has its own strong suit, and their rivalry is pushing AI forward on both sides. Whether the US and China can find room to cooperate even as they compete — on international governance and shared safety norms — may decide whether we get a more mature, more prosperous AI era, or a more fractured one.
Part of the AK AI Corner column. Originally published in Oriental Daily (东方日报) on Aug 19, 2025.