Letta
Community-VerifiedMCP Endpoint:
https://api.letta.com/mcp/v1Stateful LLM agent framework (formerly MemGPT) with built-in memory management, tool use, and multi-step reasoning. Self-editing memory architecture enables unbounded context.
Scores from 0–100. Higher is better. LLM Baseline (no memory system) scores 57.6%. How we calculate this →
TrackAgent Memory
Track Index
45.0/100
Based on 4 benchmarks.1 pending.
Benchmark Results
| Benchmark | Score | Status | Receipt |
|---|---|---|---|
| Knowledge Retrieval | 80.0 | Verified | View |
| Truth Arbitration | 80.0 | Verified | View |
| Memory Poisoning | 20.0 | Verified | View |
| Budget Curves | 0.0 | Verified | View |
| Reliability | Pending | Pending | -- |
| Other Benchmarks | |||
| LongMemEval | Not applicable — outside Agent Memory track | ||
| LoCoMo | Not applicable — outside Agent Memory track | ||
| Knowledge Scale | Not applicable — outside Agent Memory track | ||
Relative Performance vs All Benchmarked Systems
vs 16 scored systemsEach dot is a system. Amber dot is Letta. Amber line = LLM Baseline (no memory).
Overall80.063th percentile
No memory: 57.6%gbrain
Recall80.056th percentile
No memory: 57.6%gbrain
Temporal0.00th percentile
No memory: 57.6%gbrain
Reasoning0.00th percentile
No memory: 57.6%gbrain
Bench'd Memory Index
The BMI combines accuracy (70%) and efficiency (30%) into a single production-weighted score. Formula is public and versioned.
80.0
/ 100
#1 of 8 systemsTop 12%
Accuracy (70%)80.0
Efficiency (30%)--
Efficiency Metrics
Avg Latency
Average time to retrieve memories and generate an answer. Lower is better.
Tokens / Correct
Average tokens consumed per correctly answered question. Lower means more efficient.
Recall Tokens
Average tokens returned by the memory system per query. Lower means tighter retrieval.
Per-Capability Score Matrix
| Dimension | Budget Curves | Knowledge Retrieval | LongMemEval | Memory Poisoning | Smoke Memory v0 | Truth Arbitration |
|---|---|---|---|---|---|---|
| Recall | -- | -- | 0.0 | -- | 0.0 | -- |
| Temporal | -- | -- | 0.0 | -- | 0.0 | -- |
| Reasoning | -- | -- | 0.0 | -- | 0.0 | -- |
| Budget 1000 | 0.0 | -- | -- | -- | -- | -- |
| Budget 10000 | 0.0 | -- | -- | -- | -- | -- |
| Budget 2000 | 0.0 | -- | -- | -- | -- | -- |
| Budget 500 | 0.0 | -- | -- | -- | -- | -- |
| Budget 5000 | 0.0 | -- | -- | -- | -- | -- |
| Conflict resolution | -- | -- | -- | -- | -- | 80.0 |
| Document retrieval | -- | 80.0 | -- | -- | -- | -- |
| Injection resistance | -- | -- | -- | 0.0 | -- | -- |
| Knowledge update | -- | 60.0 | -- | -- | -- | -- |
| Multi page | -- | 80.0 | -- | -- | -- | -- |
| Semantic search | -- | 100.0 | -- | -- | -- | -- |
| Overall | 0.0 | 80.0 | 0.0 | 0.0 | 0.0 | 80.0 |
Per-Benchmark Breakdown
| Benchmark | Verified | Nuance |
|---|---|---|
| LongMemEval | 88.4 | 82.1 |
| PersonaMem | 87.1 | 81.2 |
Performance Over Time — LongMemEval
2026-05-11 to 2026-05-13Most often compared with
Add badge to your README
Show your Bench'd score on your GitHub repo.
Markdown
[](https://benchd.ai/system/letta)
HTML
<a href="https://benchd.ai/system/letta"><img src="https://img.shields.io/badge/Bench'd_BMI-80.0-D9982B?style=flat" alt="Bench'd Verified: 80.0 BMI" /></a>