Groot MoE series spans 19B-280B; top model scores 63.47 on SWE-bench Pro
AdinaYakup · x · 2026-10-27
A community member's Groot series of 4 MoE models (19B-280B) is drawing attention:
- 19B/A3B: 128K context on a 16GB GPU, reportedly beating larger models
- 65B/A6B: 28% fewer output tokens than Qwen3.8 Flash Next
- 238B/A18B: beats larger models on 6 agentic benchmarks, 1M context
- 280B/A16B: 63.47 on SWE-bench Pro, $0.13/task on DeepSWE 1.1, adjustable reasoning effort
A comparison chart shows Groot models leading Qwen3.x, Gemma-4, GLM-5.3-Flash, and DeepSeek-V4 on 16-task average score at smaller parameter counts (self-reported, unverified).
Related event: GrootTeam Releases Open MoE Models from 19B to 280B(2 posts)→
More from Models
- Alibaba Launches Qwen3-Max With Over 1 Trillion Parameters, 36T Training Tokens — emmanuelvivier · 2026-10-27
- Meta ships Muse Wonder, its first frontier model, claiming GPT-6 Astra-level scores — emmanuelvivier · 2026-10-27
- 278M Telugu model Vemana trained from scratch for ~$69 beats sub-1B open baselines — pranavmarla · 2026-10-27
- DeepSeek site shows new v4-flash fingerprints as community cracks checkpoint names — teortaxesTex · 2026-10-27
- GPU-poor dev: Opus 5.5 runs PPO for 10 minutes, gives up, and writes an LQR instead — _Stocko_ · 2026-10-27
- Mathematicians urge AI labs to stop giving designated mathematicians early access to model results — AndrewCurran_ · 2026-10-27