Memora (Standard Inference) - Remembering: leaderboard
Long-term conversational memory on Memora (2026, From Recall to Forgetting): models read weeks to months of simulated conversations in which user memories are added, updated and deleted, then answer remembering questions that must use current memories and avoid obsolete ones. Standard decoding with no reasoning tokens; forgetting-aware accuracy summed over weekly, monthly and quarterly histories, 0 to 300, higher is better.
Metric: Aggregated forgetting-aware memory accuracy (0 to 300) on remembering questions (direct recall of stored preferences, activities and goals), standard decoding with no reasoning tokens, on Memora, simulated multi-session conversations of ten personas whose memories are added, updated and deleted over weeks to quarters, with the full history in context (oldest sessions truncated beyond the context window); Forgetting-Aware Memory Accuracy per question rewards valid memories and penalizes obsolete ones, judged by majority vote of GPT-4.1, Claude Haiku 4.5 and Gemini 2.5 Flash, normalized to 0-100 per duration and summed over the weekly, monthly and quarterly settings; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 4 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | GPT-5.2 | Non-reasoning | 68.63 | #190 |
| 2 | Claude Sonnet 4.5 | 68.17 | #267 | |
| 3 | Qwen 3 32B | Non-reasoning | 66.5 | #983 |
| 4 | Memora Gemini 3 Pro Preview (Non-reasoning) (checkpoint unspecified) | 59.08 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=memora-standard-inference-remembering · How It Works · Data refreshed daily, snapshot 2026-11-22.