Qwen delays open-weight releases of 122B model, widening gap with SOTA AI

The Qwen team recently unveiled a series of new models, but notably omitted larger checkpoints at 122B, 35B, 27B, and 9B parameters. Community speculation suggests these larger models performed so strongly on internal benchmarks that the team deliberately chose not to release them immediately as open weights. If true, Qwen is likely waiting until they have even more capable models—potentially several months away—before making the current ones available. This pattern follows recent analyses showing open-source LLMs now lag 2–4 months behind proprietary state-of-the-art systems from companies like OpenAI and Anthropic.

For users running on consumer-grade hardware (e.g., a single RTX 3090 or 4090), Qwen models currently offer the best performance-to-hardware ratio among open alternatives. But with each new generation taking 1–2 months longer to drop as open weights, the competitive edge erodes. The risk is a repeat of the Meta-Llama scenario, where a dominant open-source provider eventually falls behind closed models. If Qwen continues this cadence, the open-source ecosystem could enter a prolonged drought, forcing professionals to either upgrade to expensive hardware or rely on API-based services—undermining the very premise of local AI ownership.