AI Plan Endurance Bench

Measure real token consumption and stability of AI coding plans over an 8–10 hour dev day

Problem

Users complain that Claude Code + Opus runs out of tokens, MiniMax Pro works daily but not on MAX, Kimi is expensive, DeepSeek is strong but its consumption is unknown. Everyone relies on scattered free credits or self-paid plans, with no unified benchmark for which combo lasts a full dev day.

Solution

Provide a reproducible benchmark harness that runs a set of typical dev tasks (read code, fix bugs, write tests, refactor) to measure token consumption, completion time, failure rate, and interruptions across models and plans, outputting comparable reports.

Users

Developers using AI coding tools more than 4 hours a day and tech leads of small teams

MVP

A test suite of 3–5 typical coding tasks; API configuration for OpenAI/Anthropic/MiniMax/DeepSeek; record token usage and time; generate comparison tables and recommendations

Revenue

Free basic report plus a subscription for the full benchmark library, regular updates, and team comparison views

Why now

AI coding plans are numerous and prices shift fast, leaving users without neutral measured data; an indie developer can iterate the task set quickly