aquiffoo/
aquif-3.0-preview-8b

78 5 months ago

Our biggest and smartest model, with GPT-4o-level performance.

tools

Models

View all →

1 model

aquif-3.0-preview-8b:latest

4.9GB · 128K context window · Text · 5 months ago

Readme

aquif-3.0 Preview (8B)

aquif-3.0 Preview (8B) is the latest iteration in the aquif language model series — a general-purpose LLM designed for speed, high-context processing, and real-world versatility. Built with 8 billion parameters and a 128K context window, this early release already rivals top open models.

Model Overview

  • Name: aquif-3.0-preview-8b
  • Parameters: 8 Billion
  • Context Window: 128,000 tokens
  • Architecture: Decoder-only transformer
  • Type: General-purpose LLM
  • Hosted on: Ollama

Features

  • 128K token context window for deep document and conversation understanding
  • Strong general performance in reasoning, coding, and language
  • Optimized for fast inference on local machines
  • Open and offline-friendly, ideal for edge use cases

Performance Benchmarks

aquif-3.0 Preview (8B) demonstrates significant improvements over comparable open-weight 8B models across a variety of benchmarks.

Benchmark aquif-3.0 Preview (8B) DeepSeek R1 8B aquif-3.0 Preview 2B GPT-4o Claude 3.5 Sonnet
MMLU 67 57 45 89 89
GSM8K 82 72 67 91 96
HumanEval 89 68 80 90 92

These results are approximate estimates for preview purposes only. Full evaluation suite will be released alongside the final model.

Use Cases

  • Long-document understanding (PDFs, meeting transcripts, research papers)
  • Code generation, refactoring, and explanation
  • Reasoning-heavy applications like agents and planners
  • Assistant-style chatbots and task runners

Getting Started

To run via Ollama:

ollama run aquiffoo/aquif-3.0-preview-8b