Open-Weight Qwen 4 Closes In on OpenAI's Tier-1 Models in Direct Testing

Alibaba's unreleased Qwen 4 model is proving competitive with OpenAI's premium GPT-6.1 Sol in live application tests, narrowing a gap that seemed insurmountable months ago. In a side-by-side comparison where both models were tasked with building an interactive drone racing game, independent testers reported that Qwen 4—still in training and accessed through Arena under the codename Kalmia—produced a more playable result, while GPT-6.1 Sol generated a more polished visual presentation. The finding underscores a widening crack in the performance hierarchy that has long favored closed proprietary models.