Enterprise AI inference costs hit 2026 low on price war

Enterprise AI inference costs have plummeted to a 2026 low, with average prices ranging between $1.16 and $1.18 per million tokens from August 6-8, according to research by investment bank Jefferies. This marks a 43% drop from $2.04 in late May and reflects a broader price war among major AI providers. The decline is attributed to aggressive pricing by US leaders like OpenAI, which slashed GPT-5.6 rates by up to 80% last month, and cost-efficient alternatives from Chinese open-source ecosystems, including models from DeepSeek.

The trend aligns with increased cost efficiencies across both US and Chinese tech sectors. Jefferies analysts noted that Anthropic’s Claude Opus 5 delivers performance comparable to its flagship Fable 5 model at half the price, further intensifying competition. Silicon Data’s index, tracking pricing across API providers and open-weight inference platforms, highlights how Chinese open-source models are pushing affordability boundaries. This shift benefits businesses seeking to scale AI deployments without proportionally increasing costs.