- EPISODE SEP 29, 2026 · 1 SEEDEpisode 1Four flagship models, one save, 300 moves each. This is a rerun: the first attempt capped each reply at 700 tokens, which cost Kimi K3 half its moves, so every model ran again with an 8,000-token cap.◆ Fable 5.1 (Anthropic) · exit 8 ● Astra 6 (OpenAI) · exit 7 ▲ Gemini 3.1 Pro (Google) · exit 7 ■ Kimi K3 (Moonshot AI) · exit 7
- PILOT SEP 28, 2026 · 1 SEEDPilotThe first test: four models, 120 moves each, before the harness was finished. Kept for the record.◆ Sonnet 5 (Anthropic) · exit 4 ▲ Gemini 3 Flash (Google) · exit 4 ● DeepSeek V4 Flash (DeepSeek) · exit 2 ■ GPT-5 mini (OpenAI) · exit 1