OpenRouter ranks leading coding models by developer usage

OpenRouter’s coding model rankings, updated July 2026, compare LLMs by real developer usage across code generation, debugging, refactoring, and AI coding assistant workflows. The leaderboard places Xiaomi’s Mimo V2.5 first with 7.54T tokens and 29.1% usage share, followed by MiniMax M3 with 2.79T and 10.8%, Tencent’s Hy3 (free) with 2.52T and 9.7%, Z.ai’s GLM 5.2 with 2.25T and 8.7%, and NVIDIA’s Nemotron 3 Ultra 550B A55B (free) with 1.98T and 7.6%.

The collection emphasizes long-context, agentic coding systems. Several leading models support a 1M-token context window, including MiMo-V2.5, MiniMax-M3, GLM 5.2, DeepSeek V4 Flash, DeepSeek V4 Pro, and Claude Opus 4.8. DeepSeek V4 Flash is positioned for fast inference and high-throughput workloads, while DeepSeek V4 Pro targets more complex tasks such as full-codebase analysis, multi-step automation, and large-scale information synthesis.

Model profiles highlight a mix of multimodal inputs, selectable reasoning levels, and cost-conscious inference. Hy3 offers a 256K context window and configurable reasoning modes, Claude Opus 4.8 is framed for autonomous agents and long-running project orchestration, and Step 3.7 Flash combines image and video understanding with selectable reasoning levels for coding, structured outputs, and long-context productivity tasks.