Long-term Memory on LongMemEval
90.8ScoreGemma-4-31B-it
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemma-4-31B-itBackbone=Gemma-4-31B-it, Quantization Method=BF162026.05 | 90.8 | |
| Mix-QuantBackbone=Gemma-4-31B-it, Quantization Method=Mix-Quant2026.05 | 90.4 | |
| Gemma-4-31B-it-NVFP4Backbone=Gemma-4-31B-it, Quantization Method=NVFP4 W4A42026.05 | 89.2 | |
| Qwen3.5-9BBackbone=Qwen3.5-9B, Quantization Method=BF162026.05 | 86.27 | |
| Mix-QuantBackbone=Qwen3.5-9B, Quantization Method=Mix-Quant2026.05 | 84.27 | |
| Gemma-4-26B-A4B-itBackbone=Gemma-4-26B-A4B-it, Quantization Method=BF162026.05 | 80.5 | |
| Qwen3.5-9B-NVFP4Backbone=Qwen3.5-9B, Quantization Method=NVFP4 W4A42026.05 | 78 | |
| PERSONATREEBackbone=Gemini 3 Flash2026.06 | 76.2 | |
| TiMemBackbone=GPT-5.4 Mini2026.06 | 74.9 | |
| Benchmark strongBackbone=GPT-5.4 Mini2026.06 | 74.5 | |
| PERSONATREEBackbone=GPT-5.4 Mini2026.06 | 74.3 | |
| TiMemBackbone=Gemini 3 Flash2026.06 | 73.6 | |
| Benchmark strongBackbone=Gemini 3 Flash2026.06 | 73.4 | |
| TiMemBackbone=Qwen3-32B2026.06 | 72.7 | |
| Mix-QuantBackbone=Gemma-4-26B-A4B-it, Quantization Method=Mix-Quant2026.05 | 72.45 | |
| PERSONATREEBackbone=Qwen3-32B2026.06 | 72.3 | |
| A-MEMBackbone=GPT-5.4 Mini2026.06 | 72.3 | |
| Benchmark strongBackbone=Qwen3-32B2026.06 | 72.1 | |
| Mem0Backbone=GPT-5.4 Mini2026.06 | 71.7 | |
| A-MEMBackbone=Gemini 3 Flash2026.06 | 70.9 | |
| Flat retrievalBackbone=GPT-5.4 Mini2026.06 | 70.3 | |
| A-MEMBackbone=Qwen3-32B2026.06 | 70 | |
| Mem0Backbone=Gemini 3 Flash2026.06 | 70 | |
| Flat retrievalBackbone=Gemini 3 Flash2026.06 | 69.1 | |
| Mem0Backbone=Qwen3-32B2026.06 | 68.6 | |
| Flat retrievalBackbone=Qwen3-32B2026.06 | 68.1 | |
| Gemma-4-26B-A4B-it-NVFP4Backbone=Gemma-4-26B-A4B-it, Quantization Method=NVFP4 W4A42026.05 | 62.42 | |
| Qwen3-8BBackbone=Qwen3-8B, Quantization Method=BF162026.05 | 57 | |
| Mix-QuantBackbone=Qwen3-8B, Quantization Method=Mix-Quant2026.05 | 54.85 | |
| Qwen3-8B-NVFP4Backbone=Qwen3-8B, Quantization Method=NVFP4 W4A42026.05 | 49.82 |