GSM8K Accuracy of Low-Bit Mixtral and Qwen3 GGUFs: Dataset and Reproducibility Artifact

A mixed-license dataset and reproducibility artifact covering six exact Mixtral and Qwen3 GGUF deployments on 100 shared GSM8K items. P4R1 scored Mixtral Q4/IQ2/IQ1 at 68/46/26 and Qwen at 91/97/96. The package contains privacy-sanitized item-level outcomes, deterministic paired-statistics checks, exact model and environment provenance, replication sources, and the accompanying technical report. The experiment is corrective and exploratory, not independent; the ladders are not precision-matched, and the results establish neither an architectural cause nor Qwen-tier equivalence. Benchmark text, model responses, and weights are excluded. AI assistance: Claude (Anthropic) assisted with evaluation-harness development and experiment orchestration. Claude and Codex (OpenAI) assisted with evidence auditing, analysis and figure code, citation checking, and manuscript drafting and revision. Matthew Schwartz directed the research and is responsible for its methods, results, interpretation, claims, citations, rights, and released artifacts.