Qwen 3.8 vs 3.5 generation time comparison on RTX 3090 (Test 3, Heavy 16K profile, NoThink mode)
The chart compares execution time for three task types — a short news item, a long-form article and a photo description — across three Qwen configurations in NoThink mode. On the RTX 3090 24GB the Qwen 3.8 27B model is the slowest while the compact 9B variant is markedly faster, a key factor for building a high-throughput article generation pipeline.