CrusaderBench Makes AI Agents Play Stronghold Crusader; GPT-6.2 Sol Tops Board
Hesamation · x · 2026-11-20
Hesamation shared CrusaderBench, an independent benchmark where AI agents play Stronghold Crusader: each model views the game via screenshots, acts with mouse and keyboard, and runs a medieval village for 33-45 game minutes on each of three maps. Score anchors: 0 = doing nothing, 100 = a human player's game (80% net worth, 10% population, 10% popularity).
Key results:
- GPT-6.2 Sol leads overall: 55.7 on The Sinkhole, 60.2 without arms, 123.0 on Large Barren Desert (no arms)
- Gemini 4 Neon is the only non-GPT model matching Claude Fable 5.2 on gameplay
- Board includes GPT-6.1 Sol, Claude Opus 5.5, Kimi K3.1 Pro, DeepSeek V4.1 Flash, GLM 5.3 Flash, and more, plus a score-vs-cost Pareto frontier chart
Post-game interviews are entertaining: Claude Fable 5.1 admitted, "I never checked the actual conversion ratios; I reasoned from unit prices and got it backwards."
Related event: CrusaderBench Tests 15 Models on Stronghold Crusader, GPT-6.2 Sol Leads(2 posts)→
More from Models
- Reddit: Claude Max for $29 Instead of $200 — How Do Resellers Profit? — mehdiweb · 2026-11-20
- Gemini 4 Neon feels better-judged than Carbon, but cache hit rates and TTFT are bad — lyraxana · 2026-11-20
- LMArena publishes Peter Gostev's leaked system prompts collection on YouTube — arena · 2026-11-20
- CrusaderBench: GPT 6.2 Sol Tops Stronghold Benchmark at 79, Gemini 4 Neon Close — Hesamation · 2026-11-20
- Cloudflare's clairic, a Qwen3.5-based multimodal info-extraction model, trends on Hugging Face — Cloudflare · 2026-11-20
- Google's comeback draws 'this is a real Code Red' reaction as rivals feel the heat — iruletheworldmo · 2026-11-20