Gauntlet AI Launches TokenSwitch to Cut Frontier Model Token Costs
Reactions from ranked influencers
2 posts
Nearly everyone is spending WAY too much on AI tokens because they’re defaulting to the most powerful frontier model. This was a massive (and extremely expensive) problem for us at Gauntlet AI, but WE SOLVED IT. And we’re making it available to all. We call it TokenSwitch.
See more at http://TokenSwitch.co: It’s a one-line instillation that works natively with Claude Code, Cursor and Codex. Every prompt is instantly classified based on complexity, cost, sensitivity, and risk, and routed to the appropriate model, (including free and open source models!) We’ve found that free and open source models are now handling about HALF of all requests. You can also write your own rules/preferences that TokenSwitch will follow. If a cheap model fails, TokenSwitch can instantly “upgrade” models, and log the failing attributes for future learnings. Additionally, it lets you instantly view total token spend from ALL PROVIDERS in one place. You can view by provider, prompt type, repo, team member, etc; a layer of visibility that’s annoyingly difficult bouncing between various provider data sources and dashboards. We hope you’ll try it and give us feedback. We’ve priced it very low ($20/mo for individuals or $200/mo for your whole team.) Shouldn’t be hard to make up that delta in inference costs.