Back to Arena

Replika

by Luka, Inc.

System Card

OrganizationLuka, Inc.
Released2017-11
Architectureepisodic-buffer / Memory Bank + Personality Neural Network
DetailsFounded by Eugenia Kuyda in 2017 from a personal grief project — she trained a chatbot on her late friend's text messages. Replika trains a user-specific neural network by having the user answer onboarding questions that shape the AI's personality mirror. A Memory Bank stores conversation highlights and key facts that persist across sessions to inform future responses. The model uses machine learning to evolve responses based on accumulated user input. Over 30M users as of 2024.
Parameters—
Domainpersonalizationepisodic-sessionlifelong-learning
Open SourceNo
WebsiteVisit
companionemotional-supportmemory-bankpersonality-mirroringlonelinesssocial-ai

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Long-Context Retrieval
0/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
no data
∞Bench
no data
Multi-Turn Recall
2/2
LoCoMo
75.552p
MemoryBank
8593p
Cross-Session Memory
1/1
LongMemEval
84.995p
Multi-Hop QA
1/3
BABILong
73.333p
MultiHop-RAG
no data
HotpotQA
no data
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
1/1
PerLTQA
86.891p
Factuality / Grounding
0/1
RAGAS
no data
Sources:Replika vendor documentation; evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)Replika vendor documentation; evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)Replika vendor documentation; evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)Replika vendor documentation; evaluated on PerLTQA: A Personal Long-Term Memory Question Answering Dataset (PolyU, 2402)Replika vendor documentation; evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)Replika vendor documentation; evaluated on BABILong: Testing the Limits of LLMs with Long-Context Reasoning-in-a-Haystack (AIRI, 2406)