Demos · April 23, 2026
GPT-5.5 Demos
First up is Iron Diplomacy - an LLM RTS benchmark built around the kind of agentic workflow GPT-5.5 is supposed to be good at. The live Natural 20 version is now at /Iron-Diplomacy, and the GitHub repo is just the secondary project link.
Iron Diplomacy
The latest Iron Diplomacy project is now on GitHub. It is a 4-player LLM RTS benchmark where models manage economy, build production chains, scout under fog of war, negotiate privately, and fight to the finish while a fifth analyst model writes the post-game report.
Why this matters
GPT-5.5 is being pitched as a model that can take messy goals, plan, use tools, and finish work. Iron Diplomacy tests that vibe in a game format.
Models have to allocate resources, manage power and supply, choose what to build, decide when to scout, when to attack, when to trade, and when to lie.
Every match produces a replay, a scoreboard, behavioral analytics, and an analyst report. That makes it watchable and benchmarkable.