Interactive Fiction on Jericho OOD (held-out games)
11.2Average Score (Balances)META-TTL
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| META-TTLArchitecture=GPT-5 (as Meta-Agent), Zero-shot generalization=true2026.04 | 11.2 | 10 | 1.8 | 7.7 | 25 | 35 | 24 | 28 | |
| META-TTLArchitecture=GLM-5 (as Meta-Agent), Zero-shot generalization=true2026.04 | 9.9 | 9.3 | 1.9 | 7 | 21 | 32 | 26 | 26 | |
| NaiveArchitecture=GPT-5 (as Meta-Agent), Zero-shot generalization=true2026.04 | 9.4 | 8.9 | 1.4 | 6.6 | 20 | 30 | 19 | 23 | |
| META-TTLArchitecture=Gemini 3 Flash (as Meta-Agent), Zero-shot generalization=true2026.04 | 8.7 | 9.7 | 2 | 6.8 | 18 | 36 | 28 | 27 | |
| NaiveArchitecture=Gemini 3 Flash (as Meta-Agent), Zero-shot generalization=true2026.04 | 7.8 | 8.3 | 1.6 | 5.9 | 16 | 30 | 22 | 23 | |
| Memory AgentArchitecture=Single-Agent Baselines, Zero-shot generalization=true2026.04 | 7.7 | 8 | 1.8 | 5.8 | 15 | 29 | 24 | 23 | |
| NaiveArchitecture=GLM-5 (as Meta-Agent), Zero-shot generalization=true2026.04 | 7.7 | 8.9 | 1.7 | 6.1 | 15 | 31 | 24 | 23 | |
| ReflexionArchitecture=Single-Agent Baselines, Zero-shot generalization=true2026.04 | 7.4 | 8.9 | 2 | 6.1 | 14 | 32 | 28 | 25 | |
| StaticArchitecture=Single-Agent Baselines, Zero-shot generalization=true2026.04 | 7 | 4 | 1.8 | 4.3 | 13 | 13 | 24 | 17 |