← Analyses

Product × benchmark coverage

The companion-and-pivot to benchmark coverage. That view is benchmark-centric — rows are benchmarks, the lens is "is this benchmark adopted?". This view flips the matrix to product-centric: each row is a product, each column is a benchmark, and the cell colour is the integrity tier of the strongest available citation. The flip surfaces the angle no comparable catalog publishes: which products refuse to put a peer-reviewed score on the record at all.

119 products · 25 benchmarks shown · 169 total cells filled

91.3%

of catalogued products report zero peer-reviewed benchmark scores

Of 912 records, only 79 have any benchmark mention backed by a peer-reviewed citation (arXiv, OpenReview, ACL, IEEE, ACM, PMLR). 833 products do not — either because they don't benchmark at all, or because the only citations are vendor-self-published. The toggle below narrows the matrix to the subset that does mention benchmarks but without peer-reviewed backing.

Sort:
Category:
Cell colour: Peer-reviewed Independently verified Vendor-claimed Disputed Unverifiable No mention
ProductNLongMemEval 19LoCoMo 31BABILong 5ConvoMem 2RULER 3MemoryAgentBench 2NIAH 7MMLU 17GAIA 13SWE-bench 12HotpotQA 10ALFWorld 9AIME 8WebArena 5NaturalQuestions 4OSWorld 4BrowseComp 3LongBench 3MT-Bench 3AppWorld 2ScienceWorld 2TriviaQA 2ImplicitMemBench 1Mind2Web 1PersonaBench 1
Agentic Memory T4 Research3+23.52pp54.31%
GAM (General Agentic Memory) T4 Research390%64.07 F1
MemAgent T4 Research395%
ShadowKV (ByteDance) T2 Research3
Anthropic Claude (foundation models) T1 Foundation3~77%~80%~88%
OpenAI GPT family (GPT-5 / GPT-4o / o3 / o4) T1 Foundation3~94%~86%
Supermemory T1 Memory systems371.43%
Mem0 T1 Memory systems3~26%91.630
A-MEM T3 Research260.6%
Agent KB T4 Research2+18.7pp+4.0pp
Agent Workflow Memory T3 Research2+51.1%+24.6%
AgentFold T4 Research236.2%
Agentic Plan Caching (APC) T3 Research2
CDMem T3 Research285.8%56.0%
DeepAgent T4 Research2
Dynamic Cheatsheet T4 Research250.0%
Engram (DeepSeek) T1 Research284.2+4.0 poi
ExpeL T3 Research259%
ExpeL T4 Research2
IterResearch T4 Research2
LiCoMemory T4 Research273.8%40%
LightMem T4 Research26.40%
LM2 T4 Research2+86.3%+5.0%
LongRAG T4 Memory systems221
MAGMA T4 Research20.700
MATTER T4 Research256.0 EM
MemoBrain T4 Research2
MemOS (MemTensor) T3 Research280%-30%
SeCom T3 Research2
SGMem T4 Research2
TiMem T4 Research276.88%75.30%
Titans (Google) T4 Research2
Alibaba Qwen 3 family T1 Foundation2~80%
Anthropic Computer Use T1 Agents2
DeepSeek R1 / V3 family T1 Foundation2~76%
Google Gemini 3 family T1 Foundation2~85%
Magnetic-One T3 Frameworks2
Mistral Large 2 / Mixtral family T1 Foundation284%~92%
xAI Grok 4 T1 Foundation2~30%~80%
ByteRover T2 Memory systems292.8%92.2%
Letta / MemGPT T1 Memory systems283.2%74.0%
OpenAI Operator T1 Agents238.1%38
ACON T4 Research1
Agent S T4 Research120.58%
AgentEvolver T4 Research1
AI PERSONA T4 Research18.29
Alita T4 Research175.15%
Alita-G T4 Research183.03%
ARMT (Associative RMT) T4 Research179.9%
Atlas T3 Memory systems1+80%
ATLAS T4 Research1+80%
BAI-LAB MemoryOS T3 Research1+48.36%
Beyond RAG for Agent Memory T4 Memory systems1
BrowserAgent T4 Research1
COLA T4 Research131.89%
D-SMART T4 Research1+10.1%
Darwin Gödel Machine T4 Research150.0%
Differentiable Search Index (DSI) T3 Research1
EM-LLM T4 Research1+40%
EMAT T3 Research144.3 EM
EverMemOS T4 Research183.0%
EVOLVE-MEM T3 Research158.3%
H2O T3 Research1
H²R T4 Research175.9%
HeLa-Mem T4 Research1
HELMET T3 Benchmarks / eval1
ImplicitMemBench T3 Benchmarks / eval164.1%
Inducing Programmatic Skills T4 Research140.4%
InfiniteBench (∞Bench) T3 Benchmarks / eval1
KAG T3 Memory systems1+33.5%
Mem-α T4 Research10.642
MEM1 T4 Research1
Memento T4 Research187.88%
MemInsight T3 Research1+34%
MemLoRA T4 Research181.3
MemMachine T2 Memory systems10.9169
MemMachine (paper) T4 Research10.9169
MemoRAG T3 Research1~62%
MemoryArena T4 Benchmarks / eval1
Memp T4 Research1
MemR³ T4 Research1
MemRL T4 Research1
MIRIX T4 Research185.4%
RAFT T3 Memory systems1
Retroformer T4 Research1
RGMem T4 Research1
RMM (Reflective Memory Management) T3 Research110%
RULER T3 Benchmarks / eval1
SkillWeaver T4 Research18%
SnapKV T3 Research1
Speculative RAG T4 Memory systems1+12.97%
SYNAPSE T4 Research12 F1
UI-TARS T2 Agents124.6%
Hindsight (Vectorize) T1 Memory systems191%
LangMem (LangChain) T2 Frameworks158.10%
01.AI Yi family T2 Foundation1024
Amazon Nova family T2 Foundation1~85%
Backboard T2 Memory systems1
ChatGPT — Codex (cloud agent) T1 Agents1
Cohere Command R+ / Command A T1 Foundation1~85%
Diffbot T1 Memory systems170.36%
GWM (Gaussian World Models) T3 Vertical1
ICLR 2026 MemAgents Workshop T2 Benchmarks / eval1
Memvid T2 Memory systems1+35%
Meta Llama 4 family T1 Foundation1~80%
Microsoft Phi-4 family T2 Foundation184.8%
OpenAI Codex CLI T1 Agents1
OpenDevin / OpenHands T2 Frameworks1
OpenHands (All Hands AI) T1 Agents1
Reka Core / Flash / Edge T2 Foundation1~83%
Wayve GAIA-2 / GAIA-3 T1 Vertical1-2
Claude Code (Anthropic) T1 Agents1~72%
Mastra Memory T2 Frameworks194.87%
Memoria (MatrixOrigin) T2 Memory systems188.78%
MemPalace T2 Research196.6%
Mistral Vibe (Remote Agents) + Mistral Medium 3.5 T1 Agents177.6%
OMEGA T2 Memory systems195.4%
SuperLocalMemory T1 Memory systems174.8%
Zep & Graphiti T1 Memory systems184%

What each integrity tier means

Peer-reviewed
Cited at a peer-reviewed venue (arXiv, OpenReview, ACL, IEEE, ACM, PMLR).
Independently verified
Cited at a non-vendor third party — neutral leaderboard or independent platform.
Vendor-claimed
Citation host matches the vendor’s own domain; vendor is making the claim.
Disputed
Another catalog row contradicts this number — in-cell dispute signal or >7-pt score divergence on the same benchmark.
Unverifiable
No citation that resolves to a host, sentinel value, or depth-floor cell.
No mention
Product makes no mention of this benchmark anywhere in the catalogued cells.
← Analyses hub Derived from cells.perf + cells.claims citations. Pivot helper in $lib/analyses/product-benchmark-matrix.ts.