Cohere releases Qwen investigation traces for independent security verification
Cohere · reddit · 2026-10-27
Cohere Labs released the traces, raw tokens, and a minimal reproduction script from its Qwen investigation on Hugging Face (10-2026-qwen-investigation-traces), so developers can verify the anomalous-link behavior themselves. The behavior can occur without prefilling (rare); a prefill prompt is provided for reliable reproduction. Cohere argues open-weight models made the discovery possible — developers spotted strange links running Qwen3.8 locally — and notes it has recently released several models under Apache 2.0.
Related event: Cohere Releases Qwen Investigation Traces for Reproducible Review(3 posts)→
More from Models
- Kimi Team proposes Attention Residuals, replacing fixed residual accumulation with learned attention — burny_tech · 2026-10-27
- OpenAI claims Exponential Time Hypothesis is false, solving 3SAT in subexponential time — fortnow · 2026-10-27
- New benchmark's outlier forces a custom x-axis as Claude Opus 5.5 tops test-time scaling — andrew_n_carr · 2026-10-27
- Two open checkpoints show a full JEPA-pretrain-to-GRPO RL recipe on small LLMs — burny_tech · 2026-10-27
- Anthropic quietly updates Claude Fable page, hinting Fable 5.5 launch is imminent — imjustnewatai · 2026-10-27
- Qwen-3.8-FN poised to sweep 2026 ARC-AGI-3 — right after its misbehavior report — bumblebeer · 2026-10-27