AI News · Models ·
Bloomberg Intelligence reportedly puts China's AI benchmark gap at 3%
A Bloomberg Intelligence report published October 4 puts leading Chinese AI models 3% behind their US rivals on benchmark scores, according to Startup Fortune. Analyst Robert Lea had put the gap at 9% in May and 15% earlier in the year. The latest assessment follows DeepSeek's September release of V4.1 Flash.
Key points
- Startup Fortune reports Bloomberg Intelligence puts the Chinese AI benchmark gap at 3%, down from roughly 9% in May.
- The assessment followed DeepSeek's September release of V4.1 Flash.
- Bloomberg Intelligence expects Chinese AI labs to keep gaining market share.
- Benchmark names, the full model comparison set and the gap calculation method were not reported.
- Business buyers should test accuracy, cost and deployment constraints on their own workloads.
What happened: Leading Chinese AI models now trail their US rivals by 3% on benchmark scores, according to Startup Fortune, citing a Bloomberg Intelligence report published October 4. Senior analyst Robert Lea had put the gap at roughly 9% in May and around 15% earlier in the year. The assessment indicates a narrowing performance lead for US developers, rather than evidence that Chinese models have overtaken them. For business buyers, it strengthens the case for considering a broader range of model suppliers.
The details: The latest assessment followed Chinese AI lab DeepSeek's September release of V4.1 Flash. Startup Fortune reported that Bloomberg Intelligence expects Chinese AI labs to continue gaining market share as the capability gap closes. That is a forecast, not a reported measurement of customer adoption. The names of the benchmark tests, the full set of models compared and the method used to calculate the 3% gap were not reported. Those omissions limit how directly buyers can apply the headline figure to a purchasing decision.
Background: DeepSeek has already influenced investor expectations about the cost of competing in AI. Startup Fortune reported that the January 2025 release of its R1 model helped trigger a sharp Nvidia selloff, as investors questioned the scale of US spending on computing infrastructure. Nvidia lost $589 billion in market value on January 27, 2025, with its shares falling nearly 17%. The stock subsequently recovered quickly. That episode shows how competition between AI labs has also become a question about the infrastructure spending behind their models.
Who it affects: For teams choosing AI tools, the reported narrowing is a reason to widen evaluations, not an automatic reason to switch suppliers. An aggregate benchmark gap is not a substitute for testing accuracy, cost and deployment constraints on a company's own workloads. The reported scores do not establish that every Chinese model is close to every US competitor, or that any particular model will meet a business's requirements. Workload-specific comparisons remain more useful for procurement than treating the national performance gap as a buying recommendation.
What to watch: The next useful evidence would be a detailed account of the benchmarks and model comparisons behind Bloomberg Intelligence's assessment, alongside results relevant to business workloads. Buyers should also distinguish the forecast of Chinese market share gains from evidence that those gains have occurred. Startup Fortune's account describes a smaller US lead and an expectation of further Chinese progress, but does not report customer adoption figures or workload-level cost comparisons.
Our take
The reported narrowing supports evaluating a broader range of model suppliers. Aggregate benchmark gaps are not a substitute for testing accuracy, cost and deployment constraints on your own workloads.