Vectara
Log inBook a demo

Vectara's models

Vectara's advanced end-to-end agentic models have pioneered enterprise-grade agent context and accuracy.

Try our models

Best-of-breed models at every stage

Agents are only as good as the models that power them, so we maintain state-of-the-art models for every step in the process.

Vectara modelYour inputs and infrastructure
  1. Query

    A question from a person or an agent

  2. BoomerangRetrieval model
    • Stateless hybrid search
    • Coarse-grained sorting of results
  3. SlingshotReranker model

    Fine-grained sorting of results

  4. MockingbirdGenerative LLM

    Knows how to form accurate answers based on results

  5. HHEMHallucination evaluation

    Detects hallucinations by checking the response against the results

  6. Your data

    Expresses meaning in words (tokens)

  7. Vector database

    Stores meaning in vectors

  8. Response + factual consistency score

Boomerang

Vectara’s advanced retrieval and embedding model delivers high-precision, contextually relevant search results

Multilingual semantic search

See announcement

Boomerang enables accurate retrieval when building agent context and supports hybrid search by mapping concepts to both vectors and keywords.

Boomerang supports hundreds of languages and dialects and uses zero-shot learning to quickly find relevant information across diverse domains.

Boomerang’s performance benchmarks

Performance details (June 2024)

Slingshot

Vectara’s reranking model reorganizes search results to prioritize highly relevant and useful information.

Reranking for relevance and variety

See documentation

The Slingshot model reorganizes search results to ensure relevance and avoid redundancy, delivering accurate outcomes across queries.

It can be combined with inputs based on your priorities, such as recency or keyword frequency, to ensure the most relevant information rises to the top.

Slingshot’s performance benchmarks

Performance details (May 2024)

Mockingbird

Vectara’s multilingual generative model is fine-tuned for factuality and structured output tasks.

Context-aware multilingual generation

Explore documentation

We optimized our Mockingbird LLM to generate highly relevant answers and cite sources for each claim.

Mockingbird handles queries and results from diverse domains and languages. It can generate everything modern AI applications need.

Mockingbird’s performance benchmarks

Performance details (July 2024)

Hughes Hallucination Evaluation Model (HHEM)

Vectara’s HHEM model is the most production-ready generative AI hallucination detector

Hallucination detection for any LLM

View documentation

The Hughes Hallucination Evaluation Model assesses LLMs for hallucinated results versus factual consistency.

The free version is on Hugging Face, with a more advanced version used in our platform's Factual Consistency Score.

Our Hallucination Evaluation Model is a first-of-its-kind initiative to proffer a commercially available and open-source model that addresses the accuracy and level of hallucination in LLMs.

Simon Hughes (1979 – 2023) avatar
Simon Hughes (1979 – 2023)ML Engineer & AI Researcher Vectara

HHEM’s hallucination detection performance benchmarks

Performance details (August 2024)

Remembering a champion of responsible AI

The Hughes Hallucination Evaluation Model honors Dr. Simon Mark Hughes, a pioneering machine learning engineer at Vectara. His research on AI hallucination detection, before his passing in 2023, continues to shape industry efforts toward truthful and reliable AI outputs.

Create AI Agents in minutes

Vectara is the Agent OS for trusted enterprise AI: governed, grounded, auditable agents with real-time policy and factual-consistency enforcement.

Get started