AstaBrief 8B
allenai/AstaBrief_8B
Open Source · chat · open-weights
Open
Alert me on changes
Context
41K
Max output
—
Weights
Open
API $/1M
—
Modalities
text
Released
09 Feb 2026
License: apache-2.0 · allenai/AstaBrief_8B
AI summary
● machine-written
AstaBrief 8B: Qwen3 fine-tune for cited scientific report generation
AstaBrief 8B is an 8-billion-parameter open-weights model fine-tuned from Qwen3-8B by Ai2, designed to generate cited scientific reports from research questions and literature excerpts. The model produces multi-section synthesis outputs in approximately 51 seconds, positioned as a faster alternative to the Claude-powered pipeline it replaced in Asta's Fast mode. Available for download and self-hosted deployment on consumer GPUs with Apache 2.0 licensing.
What's new
- Fine-tuned from Qwen3-8B base model with identical architecture (36 layers, 32 attention heads, 4096 hidden size)
- Generates cited reports in ~51 seconds, claimed 3.5x faster than predecessor Claude pipeline
- Post-training via 47,000 examples and 6,000 preference pairs (Direct Preference Optimization)
- Fits on single 24GB GPU, serves on vLLM or Transformers
Best for
Generating cited scientific reports and multi-section research synthesesFast report generation within Asta platform or self-hosted deploymentEdge deployment on consumer-grade GPU infrastructureOpen-source, locally-controlled research report automation