Multi-Round Context Recall v2 — tests a model's ability to recall and reason about information presented across multiple rounds of conversation at varying context lengths.
This definition has not completed editorial and provenance review. Verify its metric, version, source, and evaluation configuration upstream before using it for a model-selection decision.
Specifically targets multi-round conversation memory, not just single-prompt retrieval. Tests progressive context accumulation.
Wide score range (11–93%) suggests high sensitivity to model architecture. Relatively few models evaluated.
Higher is better. Recall accuracy percentage (0–100%).
Wide score range (11–93%) suggests high sensitivity to model architecture.
| # | Model | Score |
|---|---|---|
| 1 | 93.4 | |
| 2 | 93.0 | |
| 3 | 93.0 | |
| 4 | 92.3 | |
| 5 | 91.9 | |
| 6 | 91.9 | |
| 7 | 91.5 | |
| 8 | 90.9 | |
| 9 | 90.8 | |
| 10 | 90.8 | |
| 11 | 90.5 | |
| 12 | 90.3 | |
| 13 | 90.2 | |
| 14 | 90.1 | |
| 15 | 89.9 | |
| 16 | 89.9 | |
| 17 | 89.7 | |
| 18 | 89.6 | |
| 19 | 89.5 | |
| 20 | 89.4 | |
| 21 | 89.4 | |
| 22 | 88.8 | |
| 23 | 88.8 | |
| 24 | 88.5 | |
| 25 | 87.8 | |
| 26 | 85.8 | |
| 27 | 85.4 | |
| 28 | 85.3 | |
| 29 | 85.3 | |
| 30 | 85.3 | |
| 31 | 85.2 | |
| 32 | 85.1 | |
| 33 | 85.0 | |
| 34 | 84.9 | |
| 35 | 84.9 | |
| 36 | 84.6 | |
| 37 | 84.5 | |
| 38 | 84.1 | |
| 39 | 84.0 | |
| 40 | 84.0 | |
| 41 | 84.0 | |
| 42 | 83.9 | |
| 43 | 83.8 | |
| 44 | 83.7 | |
| 45 | 83.6 | |
| 46 | 83.5 | |
| 47 | 83.4 | |
| 48 | 83.1 | |
| 49 | 83.1 | |
| 50 | Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning) source benchpress Evidence observed Apr 13, 2026 | 83.0 |