Letta Memory Benchmark - Core Memory Update: leaderboard

The Letta agent must update a fact held in its core memory block, over 100 items repeated 3 times; the score is the share done correctly. It differs from the core memory read task in requiring an edit through the memory tool rather than an answer from context.

Metric: Score (%) on 100 core-memory update items, 3 repeats, Letta agent framework. Source: raw.githubusercontent.com. Saturation forecast: Estimated already saturated. 18 models tracked. Updated 16 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Sonnet 4 (20250514)83.67#388
2Claude 3.7 Sonnet (20250219)76.33#414
3GPT-4o (2024-08-06)73#736
4Claude Sonnet 4Thinking72.33#406
5GPT-4o Mini (2024-07-18)70.67#1511
6O4 Mini66.33#356
7Gemini 2.5 Flash (Preview 05-20)61.33#423
8GPT-4.1 Mini51.33#781
9Gemini 2.5 Flash (Preview 04-17)49.33#458
10Llama 4 Maverick Instruct FP846.67#1010
11Gemini 2.5 Pro (Preview 05-06)46.67#279
12Llama 3.3 70B Instruct44.33#1379
13Claude 3.5 Haiku (20241022)28.33#1601
14O3 Mini16.67#544
15Llama 4 Scout Instruct13.67#1579

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=letta-memory-benchmark-core-memory-update · How It Works · Data refreshed daily, snapshot 2026-11-16.