Skip to main content

System status

Coverage is stale.

Collection is paused. Latest public event: Aug 21, 2026 (11 days ago).

← Intel index

This coverage is stale.

Last updated May 13, 2026 (about 4 months ago).

product · May 13, 2026

Mem0 Releases Updated Memory Algorithm Achieving New Benchmark Highs in April 2026

Share the canonical public link.

Share as image

Mem0 launched an updated memory algorithm on April 30, 2026, featuring single-pass ADD-only extraction, agent-generated facts as first-class memories, entity linking across memories, and multi-signal retrieval combining semantic similarity, BM25, and entity matching. The algorithm boosts LoCoMo accuracy from 71.4 to 91.6 and LongMemEval from 67.8 to 93.4, while introducing BEAM-1M at 64.1 and BEAM-10M at 48.6, all with mean token usage of 6.7K-7.0K and p50 latency under 1.1 seconds. LongMemEval subcategory scores reach 94.3 for single-session user interactions and 100.0 for knowledge updates. Token efficiency stands 3-4x lower than full-context baselines using over 25,000 tokens.

The body justifies the 'notable' significance score moderately well, as it details a specific product update with benchmark improvements (e.g., LoCoMo from 71.4 to 91.6), qualifying as a noteworthy technical advancement but not a major company-defining event like a funding round or acquisition. Information density is high, with verifiable facts including the company name (Mem0), launch date (April 30, 2026), multiple benchmark scores (e.g., LongMemEval from 67.8 to 93.4), and metrics like token usage (6.7K-7.0K) and latency (under 1.1 seconds). The analysis field is explicitly '(none)', providing no strategic insight beyond the headline, so it scores zero.

Supporting evidence