Qwen3.5 feels ready for production use - Never been this excited

A developer's extensive testing of Alibaba's Qwen3.5-35B-A3B-UD-Q6_K_XL model reveals surprisingly production-ready coding capabilities that could disrupt the current AI development landscape. The model, running at 47.71 tokens/second across two GPUs (and hitting 80tps on a single GPU), successfully handled five real-world JavaScript, Go, and Rust client projects with only minor "5-minute tweak" bugs remaining. The tester compares the experience directly to using Anthropic's Claude Sonnet 4, noting particularly strong performance across the JavaScript ecosystem. This breakthrough suggests local models may have reached a tipping point where they can handle substantial professional development work previously reserved for expensive API-based models like Claude.

The implications are significant for developer economics and hardware investment decisions. The tester, who has spent $2,000 on Claude Pro Max subscriptions since June 2025 and projects $6,800 in total subscription costs through 2027, is now seriously considering investing in an RTX 6000 Pro workstation instead. This represents a potential shift toward a hybrid model where developers use APIs only for state-of-the-art spec generation and code reviews, while running most actual coding work locally. As local models like Qwen3.5 continue improving at this pace, the $200/month per developer subscription model faces increasing pressure, potentially accelerating the trend toward local AI deployment in professional development environments.