inference-net/OSSAS-Qwen3-14B

TEXT GENERATIONPricing:Input $0.48 / Output $0.96Concurrent Unit Cost:1Model Size:14BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 29, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

inference-net/OSSAS-Qwen3-14B is a 14 billion parameter Qwen 3 model fine-tuned by inference-net in collaboration with LAION and Wynd Labs. It specializes in generating structured JSON summaries of scientific research papers, extracting key elements like methodology, results, and claims. The model supports papers up to 131K tokens and achieves 73.9% accuracy on QA evaluation, comparable to GPT-5.

Loading preview...

Overview

inference-net/OSSAS-Qwen3-14B is a specialized 14 billion parameter Qwen 3 model developed as part of Project OSSAS by inference-net, LAION, and Wynd Labs. Its primary function is to generate structured JSON summaries of scientific research papers, aiming to democratize access to scientific knowledge.

Key Capabilities

  • Structured Summarization: Extracts key research elements such as title, authors, methodology, results, claims, and limitations into a standardized JSON format.
  • Paper Classification: Classifies input text as SCIENTIFIC_TEXT, PARTIAL_SCIENTIFIC_TEXT, or NON_SCIENTIFIC_TEXT.
  • High Context Length: Supports processing research papers up to 131,072 tokens.
  • Performance: Achieves 73.9% accuracy on QA evaluation, nearly matching GPT-5 (74.6%), and demonstrates a 98% lower cost compared to closed-source alternatives.
  • Robust Training: Fine-tuned on 110,000 curated research papers, using summaries generated by frontier models like GPT-5, Claude 4.5 Sonnet, and Gemini 2.5 Pro.

Use Cases

This model is ideal for applications requiring automated, structured analysis of scientific literature. It can be used for:

  • Research Discovery: Quickly extracting core information from large volumes of scientific papers.
  • Knowledge Graph Construction: Populating databases with structured data from research articles.
  • Academic Tools: Assisting researchers in literature reviews and understanding complex papers efficiently.

Limitations

While highly capable, the model may occasionally produce subtle factual errors (hallucinations) for very fine-grained details. Its unified schema might not capture all domain-specific nuances, and summaries should be considered research aids rather than replacements for primary sources in high-stakes scenarios.

Finetunes
1 models