Qwen 3.6 Plus tops AdamBench v1.1

AdamBench shifts the focus from one-shot generation to the iterative reality of agentic workflows, where speed and reliability are as critical as raw intelligence.

• –Qwen 3.6 Plus (API) surprised reviewers with the highest quality scores, cementing its position as the premier model for complex coding tasks.
• –CoPaw-Flash 9B emerged as the "king of lightweight coding," outperforming significantly larger models in test reliability and logic retention.
• –Gemma 4 variants offer the fastest iteration cycles due to concise token generation, providing an "agentic feel" despite trailing Qwen in raw scores.
• –The benchmark is grounded in consumer hardware reality (RTX 5080), filtering results based on what actually fits in local VRAM.
• –Methodology remains focused on React/TypeScript application building, providing a practical measure of a model's "daily driver" potential.