Fetching from the wire…
01
02
03
04
05
06
07
08
Agents2026-06-15 · source-backed
Engram, a bi-temporal memory engine, scored 83.6% versus 73.2% for a full-context baseline on the 500-question LongMemEval_S benchmark, a statistically significant +10.4 points, while using ~9.6k tokens instead of 79k (arXiv 2606.09900). Roughly 8x fewer tokens and more accurate. That kills the "tradeoff" framing: aggressive retrieval and filtering is both cheaper and better than dumping whole histories in. Code, a reproducible harness, and per-question logs are released CC-BY-4.0. If you're building agent memory, this is the result to internalize.
Each link below shares sources, entities, or timing with this story.
Shared entity: Code / Same source domain / Shared topic / What happened next / Tension
Both cover Code; reported by the same outlet (arxiv.org); overlapping topics (agent, benchmark, code).
Shared entity: Roughly / Same source domain / Shared topic / What happened next
Both cover Roughly; reported by the same outlet (arxiv.org); overlapping topics (agent, beat, context, token).
Shared entity: Code / Same source domain / Shared topic / What happened next
Both cover Code; reported by the same outlet (arxiv.org); overlapping topics (agent, benchmark, code, context).
Both cover Code; reported by the same outlet (arxiv.org); overlapping topics (agent, code, context, token).
Shared entity: Code / Same source domain / Shared topic / What happened next / Downstream implication
Both cover Code; reported by the same outlet (arxiv.org); overlapping topics (baseline, code).
Shared entity: Code / Same source domain / Shared topic / What happened next
Both cover Code; reported by the same outlet (arxiv.org); overlapping topics (agent, better, code).
Both cover Code; reported by the same outlet (arxiv.org); overlapping topics (agent, beat, code).
Both cover Code; reported by the same outlet (arxiv.org); overlapping topics (benchmark, code, token).