Benchmarks · creative
PingPong English (unadjusted mean)
Role-playing benchmark where LLM interrogators emulate users in multi-turn character conversations, and LLM judges rate in-character, entertaining, and fluent qualities. Maintained by Ilya Gusev. Score: Unadjusted Avg score with character:entertain:fluency weights 1:1:1.
26 models · 26 results · data as of 2026-10-29 · ilyagusev.github.io ↗
Leaderboard
Llama 31
All results
26 rows| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Llama 31 | 4.8 | |
| 2 | DeepSeek V3 | none | 4.8 |
| 3 | Opus 3 | none | 4.7 |
| 4 | GPT-4o | high | 4.7 |
| 5 | Gemini 2 | 4.7 | |
| 6 | GPT-45 | 4.7 | |
| 7 | Llama 31 405b | 4.7 | |
| 8 | Llama 31 70b | 4.7 | |
| 9 | Llama 3 70b | none | 4.6 |
| 10 | Llama 33 70b | 4.6 | |
| 11 | Grok 2 | none | 4.6 |
| 12 | GPT-4o mini | high | 4.6 |
| 13 | Mistral | none | 4.6 |
| 14 | Llama 3 | 4.5 | |
| 15 | Gemma 2 27B IT | 4.5 | |
| 16 | Llama 31 8b | 4.5 | |
| 17 | Qwen 25 72B | 4.5 | |
| 18 | Gemma 2 27B | 4.5 | |
| 19 | Qwen 2 72B | none | 4.5 |
| 20 | Gemma 2 9B IT | 4.5 | |
| 21 | Mistral Large | none | 4.4 |
| 22 | Mistral Small | none | 4.4 |
| 23 | Haiku 3 | medium | 4.4 |
| 24 | DeepSeek V2 | 4.3 | |
| 25 | Qwen 2 7B | 4.1 | |
| 26 | Qwen 25 14B | 4.0 |