GestaltLabs/Ornstein-3.5-9B-V1.5

VISIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 17, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

GestaltLabs/Ornstein-3.5-9B-V1.5 is a 9.65 billion parameter language model developed by DJLougen / GestaltLabs, fine-tuned from Qwen 3.5 9B with a 32K context length. This model is specifically optimized for reasoning and technical problem-solving, demonstrating significant gains in multi-step and graduate-level scientific reasoning. It is intended for AI research assistance and complex analytical tasks, shaping how the model processes information rather than just recalling facts.

Loading preview...

Ornstein 3.5 9B — V1.5 Overview

GestaltLabs/Ornstein-3.5-9B-V1.5 is a 9.65 billion parameter model, fine-tuned by DJLougen / GestaltLabs from the Qwen 3.5 9B base model. It is part of the Ornstein series, which focuses on reasoning and agent-oriented capabilities through a custom, automated data curation pipeline ensuring high structural quality in training data. This V1.5 release is a refined supervised fine-tune, building upon an initial reasoning fine-tune and serving as a foundation for an upcoming V2 release that will incorporate reinforcement learning methods.

Key Capabilities & Performance

Ornstein V1.5 is designed to instill a disciplined reasoning behavior, emphasizing evidence-based analysis, alternative weighing, and verification over superficial chain-of-thought. Benchmarks on the Gestalt Benchmark Suite (GBS, STANDARD-200) show substantial improvements:

  • Overall accuracy increased by +12.5 points compared to the base Qwen3.5-9B.
  • Reasoning score improved from 0.68 to 0.90.
  • GPQA (graduate-level science) score jumped from 0.36 to 0.80.
  • Coding ability is preserved, with a slight increase from 0.77 to 0.80.

Intended Use Cases

This model is particularly well-suited for:

  • Reasoning-heavy tasks.
  • AI-research assistance.
  • Technical and scientific problem-solving.
  • General conversational applications requiring analytical depth.