Miles per gallon for coding agents.
Every plan sells you a tank. Vendors won’t tell you the mileage, so your agent runs dry mid-refactor. AgentMPG combines official caps with crowd-reported burn rates from real builders to rate every major coding-agent plan by how far each dollar actually gets you.
Value leaderboard
Best by budget →Full details →Value score = AgentMPG editorial estimate of agentic work per dollar, 0–100 (not a measurement). How we rank.
Official limits
We read the fine print on 5-hour windows, weekly caps and usage pools so you don't have to.
Crowd burn rates
Anonymous reports of how fast real workloads hit the wall — refactors, greenfield, parallel agents.
One number
A value score that blends both, updated as plans and limits shift.
FAQ
Which coding agent plan is the best value?
On AgentMPG's current leaderboard, the top estimated value scores are Claude Max (was Max 5x) ($100/mo, 88/100), Claude Ultra (was Max 20x) ($200/mo, 86/100), ChatGPT Pro 200 ($200/mo, 80/100). Value scores are editorial estimates of agentic work per dollar, blended from official limits and crowd burn-rate reports; they are not measurements. The best plan for you depends on your workload and which models you need.
What is the best AI coding plan for about $20 a month?
Among plans at $25/mo or less, ChatGPT Plus ($20/mo) has the highest estimated value score (80/100). Claude Pro, Google AI Pro, Windsurf Pro and Copilot Pro are the other low-cost options we track.
Is Claude Max or ChatGPT Pro better for coding agents?
Both have $100 and $200 tiers. Claude Max uses rolling 5-hour sessions plus weekly limits shared with Claude chat; ChatGPT Pro has no 5-hour Codex limit but has weekly limits, and OpenAI cut the allowance for new Pro subscriptions from Sep 29, 2026. OpenAI does not publish exact Pro allowances, so its scores carry extra uncertainty.
What changed with Cursor pricing?
Cursor Pro was replaced by Cursor Premium at $30/mo, alongside Super ($100) and Ultra ($200). The usage pools cover Grok models; third-party models like Claude and GPT are billed separately at API list price.
How is the AgentMPG value score calculated?
Each plan's score (0–100) is an editorial estimate combining official vendor limits, checked against the vendor's pricing page on the date shown, with anonymous crowd reports of how fast real workloads hit those limits. Plans without crowd data are marked as provisional.
How often are prices and limits verified?
Every plan lists its official source link and the date it was last verified. Limits change often, so check the vendor page before buying.