A post by AI researcher ℏεsam (@Hesamation) on X at 5:34 p.m. KST on August 20, 2026 is getting attention. Bloomberg laid out the state of the US-China AI race in a graphic titled "US Lead in the AI Race With China Is Rapidly Narrowing," and in it, Moonshot AI's Kimi K3 showed performance close to Anthropic's Claude Fable 5.
The gap really is shrinking
According to Bloomberg Intelligence's tally, the performance gap between Chinese and US AI models narrowed from 9% in May 2026 to 6% in June, a record low. Since its July release, Kimi K3 has closed the gap with Anthropic's and OpenAI's leading models and even pulled ahead on some benchmarks. Bloomberg also reported that Alibaba's Qwen3.8-Max matched or exceeded Fable 5.
The price gap is bigger. There's comparison data that Kimi K3 is about 70% cheaper than Fable 5 per token, and about 66% lower in the cost of actually completing a task (Together AI benchmark). The "70% cheaper" Hesamation mentioned corresponds to the per-token figure.
GLM-5.3 isn't even on the list yet
The more striking point is that Z.ai's GLM-5.3 isn't yet included in this chart. GLM-5.3 was announced in August 2026, but the release of its weights has been delayed by about two weeks for a safety review. That means the gap shown in the chart doesn't yet reflect all of China's latest models.
Hesamation summed it up: "US labs still hold the frontier, but China is making it impossible to charge frontier prices." That reads as the gap closing faster on price competitiveness than on model performance.