Epoch AI data show listed inference prices at a fixed MMLU threshold fell about 99.5%, while CHINA INDUSTRIAL SECURITIES CO.,LTD. (CISC) argues broad AI deployment may become a bigger competitive advantage.
Epoch AI lists GPT-4-0314 at $37.50 per 1 million tokens on Mar. 14, 2023, with an MMLU score of 86.0. Its listing for Gemini 2.0 Flash shows $0.18 per 1 million tokens on Feb. 5, 2025, and a score of 88.0. Calculated from those figures, the decline is about 99.5%. The comparison identifies the lowest listed prices for models meeting or exceeding a set score threshold; the scores do not establish equal performance across all tasks.
The listed prices are for inference and do not show providers’ internal operating costs. Silicon Data describes its LLM Token Expenditure Index as a daily measure that weights provider prices by observed consumption. The index showed $0.88 per 1 million tokens on Nov. 2, 2026, making it a different measure from Epoch AI’s benchmark comparison.
CISC says its Research Institute of Economics and Finance made the application-focused case at a May 2026 investment strategy conference. The firm argues that gains in open-source models could intensify price competition and reduce returns from selling tokens alone.
CISC points to content, gaming, advertising, office productivity, transactions and local services as areas where AI could be integrated. It says Chinese platform companies may have advantages through their application range, user bases, industrial ecosystems and engineering capabilities. Those benefits are a thesis, not reported company-level results.
The firm lists slower-than-expected progress in AI development and commercial adoption, as well as global economic and policy uncertainty, among the risks to its view.
