Trinity Mini (free) - API, Providers, Stats
Trinity Mini is a 26B-parameter (3B active) sparse mixture-of-experts language model featuring 128 experts with 8 active per token. Engineered for eff...
Welcome! Type "help" for available commands.
$
Open Source Repository
Summary
The repository hosts a public leaderboard that measures how often large language models hallucinate when summarizing short documents. It provides model‑wise hallucination rates, factual consistency metrics, and links to interactive versions of the evaluation.
Highlights
- Evaluates LLMs on hallucination frequency during summarization of brief texts
- Reports detailed metrics such as hallucination rate, answer rate, summary length, and factual consistency scores
- Lists top‑performing models with percentages (e.g., antgroup/finix_s1_32b at 1.8%)
- Data last refreshed on December 18, 2025 and includes links to an interactive Hugging Face demo and earlier versions
- All code and data are openly available under a permissive license
auto-generated
Context
Audience
Researchers and engineers working on large language models, AI product teams focused on summarization quality, and data scientists interested in evaluating model reliability
DomainGenerative AI Evaluation
Formatopen‑source GitHub repository with markdown documentation, CSV metrics, and notebook demos
Accessfree online, open source
Topics
Related
Hugging Face demo of the leaderboardPapers on LLM factual consistencyOther Vectara evaluation datasetsModel cards for listed LLMsAI model benchmarking platforms
