Faced with surging artificial intelligence operational costs, US enterprises are rapidly shifting from high-cost closed models like OpenAI and Anthropic to open-weight models with publicly released parameters. On cloud platform Vercel’s AI Gateway, the share of tokens processed by open-weight models climbed to 56% by August 2026. Companies are increasingly routing routine tasks through affordable open models to cut expenses and prevent data leaks.
Key Takeaways
- Open-weight AI model token share on Vercel surged from 7% in December 2025 to 56% in August 2026.
- Tinder adopted open models after AI spending reached $10 million, while AT&T plans to expand its open model share to 70%.
- Anthropic used its internal AI ‘Claude Tag’ to cut service loading times from 3.1 seconds to 0.55 seconds in two weeks.
- Practical cost-saving tips for individual workflows and exact revenue loss figures for proprietary AI developers were not disclosed.
Earnings Mentions Up 6x as Vercel Token Share Surges from 7% to 56%
According to research platform AlphaSense, mentions of ‘open weight’ or ‘open source’ AI models in corporate earnings calls and investor conferences in August and September 2026 rose sixfold year-over-year.
Data from US cloud provider Vercel’s AI Gateway showed that the token share for open-weight models jumped from 7% in December 2025 to 56% in August 2026.
By releasing model weights, open-weight models allow companies to deploy and fine-tune systems on their own servers for free or at minimal expense. Running models on internal networks without third-party API calls lowers per-token costs and eliminates data leak risks. Providers including DeepSeek, Zhipu, Mistral, and Nvidia are actively competing in the open model space.
Related Reading
OpenAI and Anthropic Launch Budget AI Models, Kicking Off Price War
Anthropic Unveils Claude Sonnet 5.5 with 30% Faster Output Speed
Enterprise Adoption: Tinder, AT&T, and Beyond
Enterprises are ramping up open model adoption to curb operating costs driven up by routing routine tasks like simple queries and document summarization through expensive proprietary systems.
Current adoption and operational metrics across major enterprises include:
| Company | AI Usage Metrics and Adoption Status |
|---|---|
| Tinder (Match Group) | Adopted open-weight models for non-technical user requests after annualized AI spending surged from $1 million in January 2026 to $10 million in July |
| AT&T | Processes 45 billion tokens daily; plans to raise open model share from 40% to 70% within a year |
In addition, PNC Financial Services, C.H. Robinson, and Siemens are pursuing open model adoption and cost optimization.
Anthropic Deploys ‘Claude Tag’ to Triple Speed in Two Weeks
On September 23, 2026, Anthropic announced that deploying its internal research model ‘Claude Tag’ as an autonomous optimization agent improved core user experience speeds across its web service and desktop app by roughly three times in just two weeks.
The optimizations cut the initial loading time of the Claude web service from 3.1 seconds to 0.55 seconds, and reduced Claude Code session launch times from 0.8 seconds to 0.3 seconds.
Anthropic recorded zero outages or rollbacks across more than 3,000 automated changes by writing unit tests upfront and managing roughly 200 feature flags. Key fixes included modifying about 20 lines of code for V8 regex path optimization and resolving input field bottlenecks involving 6,900 React hooks and 900 store subscriptions.
Unverified Details
Specific practical tips for individual users to deploy cost-effective open AI models in daily work (such as setup instructions, recommended tools, or personal plan savings) were not provided in the source material.
Exact quarterly revenue losses or financial impact figures for OpenAI and Anthropic resulting from enterprise migration to open-weight models were also not disclosed.
Detailed market share percentages for specific top 10 models on the OpenRouter platform mentioned in the source material were also unconfirmed.
Related Reading
- AI Hiring Boom Hits Major Companies: 69.2% of Recruiters Prefer AI Skills
- Nationwide Free AI Service Launching in December: Telecom Features and Roadmap
- Meta Launches ‘Muse’ AI Assistant for Email Organization and Shopping