Open-Weight AI Models Seize Token Lead, but Proprietary Still Make the Money
3 hours ago
- Open-weight models processed 56% of Vercel's AI Gateway tokens in August 2025, up from 36% in July and less than 10% in December 2025.
- Despite dominating token volume, open-weight models accounted for only 14% of estimated spending due to substantially lower inference costs.
- Closed-weight tokens cost approximately 7.8 times more than open-weight tokens on average, with open-weight tokens being about 87% cheaper.
- The capability gap between leading open-weight and closed models has shrunk to about 3.3%, with open-weight models at or near parity on coding, instruction-following, and general-knowledge tasks.
- Enterprises are moving large production workloads to open-weight models, reserving costly proprietary systems only for tasks that justify the premium.
- The rise of open-weight models helped push the average price per token on the AI Gateway down by 23.2% in August 2025.
- Price-sensitive customers are stepping down within proprietary model lineups, such as from Anthropic's Fable 5 to the cheaper Opus 5, driving down costs.
- Pricing is becoming the real driver of AI model adoption, rather than the capabilities of the most impressive frontier models.