TokenHunger

TokenHunger is a benchmarking platform that runs user-provided test cases across multiple AI models, scores the answers, and ranks models by cost per correct answer. It solves the problem of choosing between expensive frontier models and cheaper alternatives by providing a single decision metric. The target audience is developers, data teams, and engineering teams in companies of all sizes who need to select cost-effective AI models for their specific tasks. It is positioned as a neutral, open-source-based tool that does not sell models, with a hosted service that manages provider keys and billing, and an open-source engine available for local use.

Key features

• Cost-per-correct-answer benchmarking
• Run cases across multiple models
• Score answers automatically
• Rank models by cost per success
• Open-source engine (costbench)
• Hosted with managed provider keys
• GitHub sign-in and credits
• MCP access for agents
• Connectors for datasets and MCP
• Free cost estimates without signup