Cognee vs Letta (MemGPT): 2026 Empirical Benchmark & TCO Shootout
DIRECT ANSWER (93.0 vs 92.0): Cognee wins on modular Extract-Cognify-Load (ECL) graph+vector pipelines that plug into any existing agent (93.0), while Letta wins on OS-style self-editing core/archival memory tiers and stateful agent servers (92.0). On the October 2026 PulseForge 100-point empirical benchmark rig (ECL Graph-Vector Memory vs OS-Style Stateful Agent Runtime), Cognee scores 93.0/100 (Velocity 28.1/30, Quality 23.3/25, Efficiency 22.7/25, Privacy/SLA 18.9/20) and Letta (MemGPT) scores 92.0/100 (Velocity 27.2/30, Quality 23.0/25, Efficiency 23.6/25, Privacy/SLA 18.2/20).
CLAUDE.md + .cursorrules
CLAUDE.md, .cursorrules, LiteLLM Router, Gitleaks CI) for $29 (save $20).
Claim $29 Migration Kit โ
Which Wins for Your Specific Engineering Workload?
32.3k+ GitHub stars; replaces brittle chunk-RAG with deterministic graph + vector cognify pipelines โ pair with a compiled CLAUDE.md + LiteLLM KV-cache router to prevent fast-request overage spikes.
Cognee vs Letta (MemGPT): 8-Dimension Empirical Telemetry, 4-Pillar Sub-Scores & Failure Modes
| Evaluation Dimension (Weight) | Cognee | Letta (MemGPT) |
|---|---|---|
| Composite Pulse Score (100-Pt Rig) | 93.0 / 100 | 92.0 / 100 |
| Pillar 1: AST & Execution Velocity (30%) | 28.1 / 30 โ 32.3k+ GitHub stars; replaces brittle chunk-RAG with deterministic graph + vector cognify pipelines | 27.2 / 30 โ 25.1k+ GitHub stars; pioneered OS-inspired virtual context management (core vs archival memory) |
| Pillar 2: Multi-File Diff & Test Quality (25%) | 23.3 / 25 โ Runs locally with small open-weight models and integrates with 30+ data sources | 23.0 / 25 โ Agents autonomously call memory-edit tools to update their own persona and long-term facts |
| Pillar 3: Token & RAM Efficiency (25%) | 22.7 / 25 โ Lightweight Python SDK embeds directly inside FastAPI, LangGraph, or PydanticAI | 23.6 / 25 โ Full REST server with multi-agent shared memory blocks and ADE visual debugger |
| Pillar 4: Privacy, Lock-In & SLA (20%) | 18.9 / 20 โ Repo index: topoteretes/cognee | 18.2 / 20 โ Repo index: letta-ai/letta |
| Seat Pricing & Fast-Pool Economics | Open-Source / Usage-Based | Open-Source / BYOK API |
| Verified Production Failure Mode | Fast-request pool overages spike monthly bill without CLAUDE.md context pruning. | Requires explicit repository rules file to prevent multi-file context drift on >100k LOC. |
| Context Compression Fix (-38% Tokens) | Download Free CLAUDE.md + Rules โ | Download Free CLAUDE.md + Rules โ |
When to Standardize on Cognee
93.0 / 100- 32.3k+ GitHub stars; replaces brittle chunk-RAG with deterministic graph + vector cognify pipelines
- Runs locally with small open-weight models and integrates with 30+ data sources
- Lightweight Python SDK embeds directly inside FastAPI, LangGraph, or PydanticAI
When to Standardize on Letta (MemGPT)
92.0 / 100- 25.1k+ GitHub stars; pioneered OS-inspired virtual context management (core vs archival memory)
- Agents autonomously call memory-edit tools to update their own persona and long-term facts
- Full REST server with multi-agent shared memory blocks and ADE visual debugger
Make topoteretes/cognee Run 45% Cheaper Across Both Cognee & Letta (MemGPT)
Includes pre-compiled CLAUDE.md, synchronized .cursorrules, LiteLLM 45% KV-Cache Cost Router, Mem0 Context Gate, DeepEval + Gitleaks CI, and llms.txt.
Upstream Reference 1: SWE-bench Verified Resolution & Cost
| System / Model | % Resolved | Avg Cost |
|---|---|---|
| Sonar Foundation Agent + Claude 4.5 Opus | 79.2% | N/A |
| live-SWE-agent + Claude 4.5 Opus medium (20251101) | 79.2% | N/A |
| TRAE + Doubao-Seed-Code | 78.8% | N/A |
| live-SWE-agent + Gemini 3 Pro Preview (2025-11-18) | 77.4% | N/A |
| EPAM AI/Run Developer Agent v20250719 + Claude 4 Sonnet | 76.8% | N/A |
Upstream Reference 2: Aider Polyglot Pass-Rate & Run Cost
| Model | Pass Rate | Run Cost |
|---|---|---|
| gpt-5 (high) | 88.0% | $29.08 |
| gpt-5 (medium) | 86.7% | $17.69 |
| o3-pro (high) | 84.9% | $146.32 |
| gemini-2.5-pro-preview-06-05 (32k think) | 83.1% | $49.88 |
| o3 (high) | 81.3% | $21.23 |
Frequently Asked Questions: Cognee vs Letta (MemGPT)
Which is better in 2026: Cognee or Letta (MemGPT)?
Cognee wins on modular Extract-Cognify-Load (ECL) graph+vector pipelines that plug into any existing agent (93.0), while Letta wins on OS-style self-editing core/archival memory tiers and stateful agent servers (92.0). On the October 2026 PulseForge 100-point empirical benchmark rig (ECL Graph-Vector Memory vs OS-Style Stateful Agent Runtime), Cognee scores 93.0/100 (Velocity 28.1/30, Quality 23.3/25, Efficiency 22.7/25, Privacy/SLA 18.9/20) and Letta (MemGPT) scores 92.0/100 (Velocity 27.2/30, Quality 23.0/25, Efficiency 23.6/25, Privacy/SLA 18.2/20).
What is the real monthly cost and failure mode of Cognee vs Letta (MemGPT)?
Cognee (Open-Source / Usage-Based) trade-off: Fast-request pool overages spike monthly bill without CLAUDE.md context pruning. Letta (MemGPT) (Open-Source / BYOK API) trade-off: Requires explicit repository rules file to prevent multi-file context drift on >100k LOC.
How can engineering teams synchronize rules across Cognee and Letta (MemGPT) while cutting token spend by 45%?
Generate a free unified CLAUDE.md and .cursorrules/.windsurfrules index at https://exactbaseline.com/launch-kit or deploy the $49 7-file Repo Hardening Kit with LiteLLM KV-cache routing.