VQV Signal
According to a post on X by @BAAIBeijing, BAAI, working with USTC and RUC, built AREX-2, a 27B autonomous research model, on Qwen3.8-27B. The model was trained for deep reflection and long-horizon execution, which the post describes as a test-time form of recursive self-improvement. BAAI says AREX-2 tops MLE-bench Lite at 81.8 and sets new records for models of 40B or smaller on BrowseComp (84.0) and HLE (52.6).
According to a post on X by @BAAIBeijing, BAAI, working with USTC and RUC, built AREX-2, a 27B autonomous research model, on Qwen3.8-27B. The model was trained for deep reflection and long-horizon execution, which the post describes as a test-time form of recursive self-improvement. BAAI says AREX-2 tops MLE-bench Lite at 81.8 and sets new records for models of 40B or smaller on BrowseComp (84.0) and HLE (52.6).
AI-assisted summary based on the listed source.
BAAI, working with USTC and RUC, built AREX-2 on Qwen3.8-27B and trained it for deep reflection and long-horizon execution as a test-time form of recursive self-improvement. BAAI says it tops MLE-bench Lite at 81.8 and sets new records for models of 40B or smaller on BrowseComp (84.0) and HLE (52.6). Original post...
Teams building agents or automated research tools may find these results relevant, since the model is aimed at long, multi-step research work. All scores are benchmark claims from BAAI itself, so the key thing to watch is whether independent evaluators reproduce the MLE-bench Lite, BrowseComp and HLE results. The 40B-or-smaller records also matter for teams weighing smaller models against larger ones.
VQV organizes public signals from inspectable sources. It does not independently verify the underlying report.
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
VQV surfaced this signal because it is recent, relevant to Open Source LLMs, connected to Scobleizer News Pulse (X).
No login, cookies, social SDKs, or automatic posting.
