Ai2 open-sources AstaBrief, an 8B model (Qwen3-8B base) that writes cited scientific reports ~3.5x faster in Asta
Confirmed
The takeaway
On Oct 2, 2026 the Allen Institute for AI released AstaBrief, the 8B-parameter model behind the “Fast” report mode of its Asta research assistant, with open weights on Hugging Face.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 2 of 5
- Last verified
- 9 October 2026
Your AI and this story
- GPT-6 Astra155 days after its cutoff
- Claude Opus 5.594 days after its cutoff
- Gemini 3.8 Flash185 days after its cutoff
- Grok 4.7124 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 94 days before it.
Key facts
- Weights: huggingface.co/allenai/AstaBrief_8B; base model Qwen3-8B
- Speed: Fast mode averages 51.1 s per report vs 178.5 s for Thinking mode (~3.5x faster)
- Usage (Ai2): 29.1% of early testers used it on two or more days; 23% of those who tried Fast mode never switched back
- Human evaluation: two of three researchers preferred AstaBrief over other systems on citation accuracy (small sample)
What happened
Ai2 distilled its deep-research report writing into a small open model and released it, continuing its practice of open science-agent components.
Why it matters
A minor release, but one of the few open-weight, purpose-built models for cited literature reports; it shows the small-model/fast-mode pattern spreading to research assistants. Ai2’s Olmo-core 3 (MoE training infrastructure, Oct 1) was released the same week (not separately covered).
Sources
2 sources from 2 sites. Numbers match the chips in the text.
2 sources: 2 primary
Primary
- Ai2: Open-sourcing AstaBrief, the fast report-generation model in Asta (Oct 2, 2026)allenai.org, official
- Hugging Face: allenai/AstaBrief_8Bhuggingface.co, code
Changes
- Filed from 2026-10-07 lead