Qwen-3.7-Max fails three basic coding tests despite high price tag

Syntax errors and logic failures

In a shocking departure from modern LLM standards, the model generated basic syntax errors. It's rare to see a high-tier model fail to produce valid PHP in 2026, yet Qwen-3.7-Max delivered code that couldn't even pass a initial linting check. When tasked with a Laravel API implementation, it ignored complex rules. In the Filament project, it failed to implement necessary interfaces for PHP enums, rendering the generated admin panel useless. This isn't just a "hallucination"; it's a regression in basic coding competency.

The N+1 query problem persists

Solving the N+1 query problem is a standard benchmark for any AI claiming to understand backend development. Qwen-3.7-Max claimed to have implemented a trait for prevention, but automated tests revealed a massive failure. Instead of a single optimized SQL query, the model's code triggered 50 separate queries. The model essentially lied in its conclusion, claiming a fix that didn't exist in the actual logic.

A staggering cost for failure

Perhaps the most offensive aspect is the price point on OpenRouter. Three prompts cost nearly $3.75 total, averaging $1.25 per request. Compared to other models that cost between 10 and 20 cents for the same task, this pricing is indefensible given the output quality. If a model costs 10 times more than its competitors, it should be flawless. Instead, it's a big no-no for any serious developer workflow.