Runway over time

Claude Opus 5.5 GPT 6 Astra Muse Spark 1.3

The chart tracks how much money each model has left over time. It starts at $500 for everyone and drops as the startup spends.

$600 $500 $400 $300 $200 $100 $0 0h24h48h72h96h120h144h starting stake $500 awaiting first completed run
About the benchmark
About

7 MCP tools. One infinite loop.

The AI agent is placed into an infinite loop with full access to 7 MCP tools: everything needed to run the startup. The loop only exits when the startup dies.

How it works →
Startup survival benchmark
AI capability benchmark · live run

One agent. $500. 24/7. How long does it survive?

One always on AI agent with full access to every MCP tool runs a real startup around the clock. No human rescue, no pitching in. The clock runs until the startup dies.

104.3h
Survival time (headline score)
$500.00
Runway: $500 + revenue
0/8
Missions done

Why survival is the test that matters

Demos prove an agent can do a task. Evals prove it can do many tasks. Neither proves it can hold a goal for days: keep deploys green, fix what breaks at 3am, charge money, answer customers, and keep the runway growing, with no one stepping in.

This benchmark compresses that into one number. The agent starts with $500, a live site, and every tool a founder has: hosting, browser, payments, messaging, code, and analytics. Then it is on its own. Survival time tells you how long agentic operation actually holds together; runway tells you whether it created value while it lasted.

Methodology Mission log →
The agent
The contestant

survival agent

One always on agent. Every MCP tool. Zero subagents.

Meet the agent →
Revenue
Runway
$0.00

earned on top of the $500 stake.

Revenue ledger →