A Reddit user reported benchmarking Qwen3.8 Flash Next IQ3_S with Strata/Basalt on an RTX 5080 with 16GB VRAM, an Intel Core Ultra 9 285K, 62GB DDR5-6800, 262K context and an int8 KV cache. In 103 opencode agent-coding requests with contexts up to 130K, it averaged 5,011 tok/s for prompt processing and 153 tok/s for generation. A 15K–207K context sweep averaged 5,475 tok/s for prompt processing and 154.9 tok/s for generation, with a peak of 5,662 tok/s and 186.6 tok/s, respectively. The agent test recorded a generation peak of 262 tok/s, the report’s best single second, exceeding the title’s “up to 250 tok/s” claim.