Hindsight vs LangMem: accuracy, cost and latency
Independent results on the same 30 questions of LongMemEval-S (cleaned, 2025-09), with the same reader model and the official judge. Pilot sample, so accuracy is shown with 95% intervals.
Verdict
Generated from the pilot data, n=30 per system- AccuracyHindsight is more accurate: 90.0% (95% CI 74–97%) vs 33.3% (95% CI 19–51%) for LangMem. The 95% intervals do not overlap, even at n=30.
- CostLangMem costs 2.8× less per 1,000 questions: $61.6 vs $172 for Hindsight.
- LatencyLangMem answers faster: median 3.5s vs 5.4s for Hindsight per question (p90 5.3s vs 8.1s).
- IngestionSimilar ingestion time: 8.8 min for Hindsight vs 8.5 min for LangMem per chat history (median).
- ContextThe reader sees a median of 18k characters of context per question with Hindsight and 23k with LangMem.
Key numbers side by side
Bold marks the better value. Accuracy rows are not bolded when the 95% intervals overlap.
| Metric | Hindsight | LangMem |
|---|---|---|
| Accuracy (official judge) | 90.0% | 33.3% |
| 95% interval | 74–97% | 19–51% |
| Panel accuracy | 86.7% | 26.7% |
| Cost per 1,000 questions | $172 | $61.6 |
| of which memory side | $171 | $61.0 |
| of which answering | $0.71 | $0.61 |
| Latency p50 | 5.4s | 3.5s |
| Latency p90 | 8.1s | 5.3s |
| Ingestion per history (median) | 8.8 min | 8.5 min |
| Context given to reader (median chars) | 18k | 23k |
Hindsight and LangMem among all tested systems
Hindsight and LangMem are highlighted; other systems are muted for context, baselines lightest. Each name links to the system's page.
Accuracy
Higher is betterShare of 30 questions answered correctly (official judge). Whiskers: 95% interval.
- Cognee90.0%74–97%
- Hindsight90.0%74–97%
- Plain RAG86.7%70–95%
- Mem083.3%66–93%
- Full context83.3%66–93%
- LangMem33.3%19–51%
Not shown: Graphiti (pilot result withdrawn, rerun in progress).
Cost per 1,000 questions
Lower is betterUSD billed (prices as of 2026-10-08). Solid: memory side (ingestion and retrieval). Light: answering.
Not shown: Graphiti (pilot result withdrawn, rerun in progress).
Latency per question
Lower is betterRetrieval plus answer, in seconds. Solid: median (p50). Light: up to p90.
- LangMem3.5sp90 5.3s
- Cognee4.0sp90 8.1s
- Mem04.3sp90 7.4s
- Full context4.8sp90 8.0s
- Plain RAG4.8sp90 7.4s
- Hindsight5.4sp90 8.1s
Not shown: Graphiti (pilot result withdrawn, rerun in progress).
Ingestion time
Lower is betterMedian time to load one question's chat history into the system.
Not shown: Graphiti (pilot result withdrawn, rerun in progress).
Accuracy by question type
Difference is Hindsight minus LangMem, in percentage points. Each type has only a few pilot questions (n), so one question can move a row by 13 points or more.
| Question type | n | Hindsight | LangMem | Difference |
|---|---|---|---|---|
| Single-session (user) | 4 | 100% | 75% | +25 pts in favour of Hindsight |
| Single-session (assistant) | 3 | 33% | 0% | +33 pts in favour of Hindsight |
| Preferences | 2 | 100% | 50% | +50 pts in favour of Hindsight |
| Multi-session | 8 | 88% | 13% | +75 pts in favour of Hindsight |
| Knowledge update | 4 | 100% | 50% | +50 pts in favour of Hindsight |
| Temporal reasoning | 8 | 100% | 38% | +62 pts in favour of Hindsight |
| Abstention | 1 | 100% | 0% | +100 pts in favour of Hindsight |
What each one is
Hindsight
Extracts facts into a memory bank on Postgres, consolidates them into observations in the background, and recalls with hybrid search plus a local reranker.
How we ran it
Embedded daemon (pg0 Postgres), one bank per question. Sessions retained with their real timestamp; recall with the question date and default budget. We wait for background consolidation to finish before asking.
LangMem
A memory manager that asks an LLM to extract, consolidate and update memories in a LangGraph store.
How we ran it
create_memory_store_manager with default instructions, one invoke() per session, LangGraph InMemoryStore with OpenAI embeddings, store.search() with the default limit (10).
What the vendors report
Hindsight
| Score | Variant | Reader | Judge | Note | Source |
|---|---|---|---|---|---|
| 90.0% | S (cleaned, 2025-09), 30-question pilot | openai/gpt-6-luna | gpt-4o-2024-08-06, official prompt | MemVerdict measurement, Hindsight 0.10.3 | This page |
| 91.4% | S | Gemini 3 Pro | GPT-OSS-120B | - | arxiv.org/abs/2512.12818 |
| 94.6% | S (cleaned) | Gemini 3.1 Pro | Gemini 2.5 Flash-Lite | - | benchmarks.hindsight.vectorize.io |
LangMem
| Score | Variant | Reader | Judge | Note | Source |
|---|---|---|---|---|---|
| 33.3% | S (cleaned, 2025-09), 30-question pilot | openai/gpt-6-luna | gpt-4o-2024-08-06, official prompt | MemVerdict measurement, LangMem 0.0.30 | This page |
| None found | - | - | - | No self-reported LongMemEval score found. | - |
Vendor-reported scores and ours are measured differently, so a gap does not by itself mean either number is wrong. Our pilot uses openai/gpt-6-luna as the reader that writes each answer, gpt-4o-2024-08-06 with the official LongMemEval judge prompt, LongMemEval-S (cleaned, 2025-09) and 30 questions, with each system set up as described above. Scores move with the reader model, the judge model and its prompt, the dataset version, the number of questions, and whether a hosted platform or the open-source package is tested.
Frequently asked questions
Is Hindsight better than LangMem?
Hindsight is more accurate: 90.0% (95% CI 74–97%) vs 33.3% (95% CI 19–51%) for LangMem. The 95% intervals do not overlap, even at n=30. With the 3-judge panel majority instead of the official judge: Hindsight 86.7%, LangMem 26.7%. The largest gap by question type is Multi-session: 7 of 8 vs 1 of 8 correct, in Hindsight's favour; each question type has only 1 to 8 questions in this pilot. These are pilot numbers (30 questions each); the full 500-question run will narrow the intervals.
Which is cheaper, Hindsight or LangMem?
LangMem costs 2.8× less per 1,000 questions: $61.6 vs $172 for Hindsight. Hindsight: $172 per 1,000 questions, of which $171 (over 99%) is memory-side (ingestion and retrieval) and $0.71 is answering. LangMem: $61.6 per 1,000 questions, of which $61.0 (99%) is memory-side (ingestion and retrieval) and $0.61 is answering. Prices as of 2026-10-08.
Which is faster, Hindsight or LangMem?
LangMem answers faster: median 3.5s vs 5.4s for Hindsight per question (p90 5.3s vs 8.1s). Similar ingestion time: 8.8 min for Hindsight vs 8.5 min for LangMem per chat history (median). Latency is measured per question as retrieval plus answering; ingestion is the time to load one question's chat history.
How were Hindsight and LangMem tested?
Both were run on the same 30 questions of LongMemEval-S (cleaned, 2025-09) as every other system, by the same harness. For each question, the system ingests that question's chat history, then retrieves context that one reader model (openai/gpt-6-luna) uses to answer with the official LongMemEval prompt. Answers are graded by the official judge (gpt-4o-2024-08-06) and cross-checked by 3 other judges; all judges agreed on 95% of graded answers. Costs are what the providers billed (prices as of 2026-10-08). Hindsight: Embedded daemon (pg0 Postgres), one bank per question. Sessions retained with their real timestamp; recall with the question date and default budget. We wait for background consolidation to finish before asking. LangMem: create_memory_store_manager with default instructions, one invoke() per session, LangGraph InMemoryStore with OpenAI embeddings, store.search() with the default limit (10).
Which should I use, Hindsight or LangMem?
If recall accuracy over long chat histories is what matters most, Hindsight did clearly better in this pilot. LangMem is cheaper, though: LangMem costs 2.8× less per 1,000 questions: $61.6 vs $172 for Hindsight. Both are MIT licensed. These are pilot numbers (30 questions each); the full 500-question run will narrow the intervals.
More comparisons
Other comparisons with Hindsight
- Cognee vs HindsightTied on accuracy (90.0% each); Cognee 1.2× cheaper
- Full context vs HindsightStatistically tied on accuracy at n=30; Full context 13× cheaper
- Graphiti vs HindsightGraphiti: pilot result withdrawn, rerun in progress
- Hindsight vs Mem0Statistically tied on accuracy at n=30; Mem0 1.8× cheaper
- Hindsight vs Plain RAGStatistically tied on accuracy at n=30; Plain RAG 31× cheaper
Other comparisons with LangMem
- Cognee vs LangMemCognee more accurate; LangMem 2.3× cheaper
- Full context vs LangMemFull context more accurate and 4.7× cheaper
- Graphiti vs LangMemGraphiti: pilot result withdrawn, rerun in progress
- LangMem vs Mem0Mem0 more accurate; LangMem 1.5× cheaper
- LangMem vs Plain RAGPlain RAG more accurate and 11× cheaper