Summary
Trellis Data's Agentaus 1 is a 220B-parameter sovereign model that, according to posts by its backers, scores 84.4% on MMLU-Pro, runs on Australian servers, does not train on user data, and needs half the GPUs of similar-size models. Santiago notes that it does not appear to be open-weight. Separately, ClementDelangue introduced Qwen3.8-27B-pi-adapted, a reinforcement-learning LoRA tuned for the pi coding agent, but the snippet cuts off before the benchmark results for that model.
AI-assisted summary based on the listed source.
What happened
Trellis Data's Agentaus 1 is a 220B-parameter sovereign model that scores 84.4% on MMLU-Pro, runs on Australian servers, does not train on user data and needs half the GPUs of similar-size models, according to posts by its backers. Santiago notes it does not appear to be open-weight. ClementDelangue introduced...
Why it matters
For builders and businesses weighing open-source options, the open-weight question matters, so the reported lack of open weights for Agentaus 1 would limit how much outside teams can inspect or self-host it. Teams using coding agents may want to watch for the full benchmark results on the Qwen adapter. The performance, GPU and data-handling claims come from backers and posts, so independent testing and official documentation are the next things to check.
Signal Intelligence
Signal Strength 82%
Technical label WATCH
Public Interest 50
Category OPEN SOURCE
Reader Depth TECHNICAL
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
Public Interest components
Recognizable Entity Score 73
Practical Impact Score 28
Novelty Interest Score 70
Consequence Score 26
Curiosity Score 16
Shareability Score 62
Why this is here
VQV surfaced this signal because it is recent, relevant to Open Source LLMs, connected to Scobleizer News Pulse (X).