Qwen3.8-flash-next IQ3_S continues to amaze me. This is Qwen3.8-flash-next-GSQ-RSO-IQ3_S, running on my 3090 + 64GB RAM Strata backend of course. Goated One-shot, no external assets, three.js island, a ring-shaped coral atoll, turquoise lagoon, swaying palms, waves breaking on the reef, a sailboat drifting around The quality of waves, the swell timing and animation is really good, the palm trees getting this good was a surprise as well, super impressed by the result. Single file output, without using any coding or agent harness, this is raw result.

Confirmed: GSQ-RCO IQ3_S for Qwen3.8 Flash Next is excellent. It’s slightly larger than UD Q2 K XL, but significantly better in both token efficiency and accuracy. I was only able to evaluate four models on agentic coding, but clearly models that performed well on non-agentic evals, such as the Q1 variants, can perform poorly on long-horizon tasks. Based on these results, I wouldn’t go below IQ3_S, in terms of size, if you expect performance reasonably close to the original model. I’ve updated my article on these GGUF models with more details: (in the chart below the model sizes exclude the ngram table)