Epoch AI Live-Tests GPT-5.6 Sol Playing Slay the Spire

Epoch AI launched the EpochAIPlays livestream event to benchmark the GPT-5.6 Sol model's agent performance in the card game Slay the Spire, featuring real-time commentary by AlephNuul. The test aimed to evaluate the model's dynamic performance in a real game scenario rather than just focusing on static scores.

Confirmed

The livestream event kicked off on Thursday. According to observations and screenshots shared by Jsevillamol, GPT-5.6 Sol demonstrated impressive capabilities in the game, smoothly clearing Acts 1 through 4 in one go and nearly completing an Act 5 run that started that morning.

Why it matters

This test serves as a highly shareable AI moment. It visually demonstrates to the public how frontier AI models can perform step-by-step logical analysis and make strategic decisions in complex, non-deterministic game environments, proving their strong potential for multi-step reasoning and agentic applications.

2026-07-22 ~ 2026-07-24 · 5 related posts

• EpochAI will stream GPT-5.6 Sol playing Slay the Spire on Thursday — Jsevillamol · 2026-07-22
• [source] GPT-5.6 Benchmarked on Slay the Spire, Shows Surprising Gaming Prowess — Jsevillamol · 2026-07-24
• Epoch AI will benchmark GPT 5.6 Sol live on Slay the Spire with commentary — Jsevillamol · 2026-07-24
• Epoch AI Live-Streams GPT-5.6 Playing Slay the Spire — Jsevillamol · 2026-07-24
• [source] GPT-5.6-Sol is shown nearly clearing an A5 Slay the Spire run — Jsevillamol · 2026-07-24