DAKERADakera on LoCoMo: 88.2%
Full methodology, category breakdown, and reproduction steps
Dakera scores 88.2% on the full LoCoMo benchmark — 50 sessions, 1,540 questions, no LLM in the retrieval path · LLM-judge scored per LoCoMo protocol. Here is every number, broken down by category, with the evaluation code so you can verify it yourself.
Four question categories
LoCoMo tests four distinct recall challenges. Dakera leads on three; temporal inference is our hardest category and an active area of improvement.
What 88.2% actually means
The benchmark dispute
Other vendors claim higher LoCoMo scores: Zep at 94.7%, Mem0 at 92.5%. Both numbers are actively disputed — Mem0 contests Zep's methodology, Zep disputes Mem0's. The disagreement is about what the benchmark actually measures.
Dakera's 88.2% uses the standard LoCoMo evaluation with no LLM post-processing — a single retrieval pass, no reranking step, scored by LLM judge as designed. Mem0's score uses LLM-assisted reranking which adds external API calls per query. These are different experiments. Dakera's number is lower but measured consistently with the baseline protocol. Full methodology is documented at the benchmark methodology post.
Dakera's LoCoMo scores — full breakdown
All numbers below are Dakera's own results on the full LoCoMo dataset (v0.11.94, June 2026). Methodology documented in full at the benchmark methodology post.
| Category | Score | Questions | Description |
|---|---|---|---|
| Overall | 88.2% | 1,540 (full dataset, 50 sessions) | Standard single-pass evaluation, no LLM reranking |
| Cat1 — Single-hop recall | 86.9% | 282 questions | Direct facts from recent or distant sessions |
| Cat2 — Multi-hop reasoning | 85.4% | 321 questions | Cross-session reasoning chains requiring multiple recall steps |
| Cat3 — Temporal inference | 73.9% | 92 questions | Time-anchored questions — our most challenging category, actively improving |
| Cat4 — Open-domain | 91.0% | 841 questions | Mixed topics and entities spanning multiple sessions |
Evaluation methodology
All scores use the full LoCoMo dataset — 50 simulated long-term conversations, 1,540 questions. We do not use sampled subsets. Standard single-pass retrieval: no LLM reranking, no post-processing synthesis step. Results are scored by LLM judge on the LoCoMo framework.
Version: Dakera v0.11.94, evaluated June 2026. Full evaluation protocol and reproducibility details are documented in the benchmark methodology post.
How we evaluate
We publish our evaluation methodology in full so you can reproduce, audit, or challenge the results.
Dataset: Full LoCoMo benchmark — 50 simulated long-term conversations, 1,540 questions across all four categories. We do not use sampled subsets. Two percentage points on a 100-question eval is statistically noise; on 1,540 questions it represents a real signal.
POST /v1/memory API. Session boundaries are preserved. No preprocessing or summarization.POST /v1/recall query. Dakera returns its top-k memories using HNSW + BM25 hybrid retrieval with temporal re-ranking.The full evaluation script and dataset ingestion pipeline is documented in our benchmark methodology post. Reproducibility is a first-class requirement — if you find a discrepancy, open an issue on GitHub.
What 88.2% means for your agent
Benchmark scores translate directly to agent behavior. Memory recall failures cause agents to repeat questions, forget context, and give inconsistent answers.
At 88.2% recall, your agent remembers what the user told it — across sessions. No "as I mentioned earlier" failures.
91.0% open-domain recall means agents carry context across days, weeks, and months — not just within a single conversation.
Sub-10ms recall at P99. No LLM rerank post-pass. Your agent gets the right memory fast enough to use it in real-time.
How Dakera compares
Other vendors publish LoCoMo results using different protocols. Here's the context.
| System | LoCoMo score | Methodology |
|---|---|---|
| Dakera v0.11.104 | 88.2% | LLM-judge accuracy · single-pass retrieval · no LLM in retrieval loop |
| Letta Filesystem | 74.0% | LLM-judge · Letta's published evaluation (letta.com/research) |
| Zep (corrected, 2025) | 75.1% | LLM-judge · Zep's own corrected score after fixing evaluation errors |
| Mem0 (Letta independent eval, 2025) | 68.5% | LLM-judge · third-party evaluation by Letta |
Mem0's self-reported 2026 score (92.5%) and Zep's 2026 claim (94.7%) use different models and evaluation setups — not directly comparable. See methodology for details.
The memory layer that proves its numbers
Dakera is self-hosted and open core. Run it in 5 minutes, reproduce these benchmark results on your own hardware, or reserve a spot on the managed cloud.
Reproduce: Full methodology · Compare: vs Mem0 · vs Zep · vs Letta · All →
Give your AI agents persistent memory
Self-host free today — or join the Dakera Cloud waitlist for managed hosting, SLA & founder pricing.