Qwen 3.5 27B FP8: benchmark results

Provider: Alibaba. Released 2026-02-24. Access: Open.

Unified ELO 1732 ± 27, rank #174 of 1605 rated models, from 67 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
EuroEval Bulgarian Common Sense Reasoning83.19Common Sense Reasoning Average Score (%)100
IFEval-TR - Average (loose)79.11Accuracy (%)100
IFEval-TR - Translated (loose)69.94Accuracy (%)100
IFEval-TR - Translated (strict)69.1Accuracy (%)100
IFEval-TR - Turkish (loose)88.27Accuracy (%)100
IFEval-TR - Turkish (strict)80Accuracy (%)100
MultiGlobeQA (Gold Triples with Code Agent)61.6Exact-match accuracy (%; oracle Tier 3: the gold triples giv100
MultiGlobeQA (KG Retrieval Agent)44.3Exact-match accuracy (%; Tier 3a: a CodeAgent writing and ex100
MultiGlobeQA (KG and Web Agent)44.1Exact-match accuracy (%; Tier 3c: a CodeAgent with both know100
EuroEval Catalan Common Sense Reasoning80.3Common Sense Reasoning Average Score (%)99.5
EuroEval Croatian Common Sense Reasoning81.04Common Sense Reasoning Average Score (%)99.5
EuroEval Danish NLU - Angry Tweets63.48Sentiment classification Score (%)99.3

Interactive version: theaggregate.ai/model?slug=qwen-3-5-27b-fp8 · How It Works · Data refreshed daily, snapshot 2026-09-26.