Vai al contenuto
AI.info

tools

HHEM (Hughes Hallucination Evaluation Model)

Open-source model that scores whether an LLM response is factually consistent with a provided source.

HHEM (Hughes Hallucination Evaluation Model)

In inglese

HHEM evaluates a source passage and an LLM-generated response, producing a score for factual consistency and hallucination detection. It is designed for developers, researchers, and teams testing summarization, question-answering, and retrieval-augmented generation systems.

The Hugging Face version can be run locally with Transformers and is licensed under Apache-2.0. Vectara also offers a more advanced commercial evaluator through its platform; enterprise deployment pricing applies there.

Features

  • Scores factual consistency between a premise and a generated response
  • Supports text-classification inference with Transformers
  • Runs locally on consumer-grade hardware
  • Provides HHEM-2.1-Open as the current open model
  • Uses an Apache-2.0 open-source license
  • Supports evaluation of summarization and question-answering outputs
  • Can be integrated into RAG evaluation workflows

Use cases

  • Score generated summaries against their source documents
  • Check whether RAG answers are grounded in retrieved context
  • Compare hallucination rates across language models
  • Evaluate question-answering responses for unsupported claims
  • Run factual-consistency checks during model testing

Pros

    Cons

      Latest updates

      Capabilities

      • Self-hosted — “HHEM-2.1-Open can be run on consumer-grade hardware” source
      • API — “HHEM-2.3 is fully integrated into Vectara and is automtically returned with every query API call.” source
      • Official SDKs — “Vectara SDK .” source
      • Runs models for you — “HHEM-2.3 is fully integrated into Vectara and is automtically returned with every query API call.” source
      • Traces and evaluates — “HHEM is a specialized model for detecting hallucinations in LLM outputs - it identifies when generated text is not supported by the provided context in RAG.” source

      Get it

      Pricing

      Starting price
      Free
      Prices checked
      2026-09-24
      Read
      by web search

      SaaS

      Price on request

      • 1 SaaS Deployment
      • Credits included with base platform package purchase

      VPC

      Price on request

      • 1 VPC Deployment (Any VPC)
      • Credits included with base platform package purchase

      On-prem

      Price on request

      • 1 On-Premise Deployment
      • Credits included with base platform package purchase

      30 Day Free Trial

      Free

      • 10,000 credits included
      • All features included for 30 days
      Official website