OpenRouter finds no accuracy gap between fp8, fp4 and fp16 for Qwen3.8 27B
AIOpenRouter tested six endpoints serving Qwen3.8 27B on October 20, 2026, and found no accuracy difference between fp16, fp8 and fp4, with all scoring 98.3% to 99.2% on 240 tasks. Lower precision was usually cheaper, with 4-bit cheaper than 8-bit for 26 of 33 models, though 8-bit was more expensive than 16-bit for 7 of 24 models served at both.
Source details
OpenRouter Blog · openrouter.aiPublished · added here
Original title: LLM Quantization: What fp8 and fp4 Do to Quality and Price