SHELFCRITTER AI Digest

MODEL

Qwen3-Coder-480B

modeltopic-notealibabaopen-source

Overview

Qwen3-Coder-480B is Alibaba’s large-scale coding-specialist model with 480B total parameters. It appears on the Terminal-Bench 2.0 leaderboard as a reference point at 23.9% via the Terminus 2 scaffold, providing the large-parameter open-weights baseline against which smaller-but-more-efficient models like Qwen3.6-35B-A3B are compared.

Timeline

  • 2026-05-17-AI-Digest — Listed on Terminal-Bench 2.0 leaderboard at 23.9% via Terminus 2 scaffold, serving as the large-open-weights reference point; scored below Qwen3.6-35B-A3B’s 24.6% via little-coder, illustrating that active-parameter efficiency can match or exceed total-parameter scale on this benchmark.

Key Developments

  1. Terminal-Bench 2.0 Reference Point: 23.9% via Terminus 2 establishes the large-parameter open-weights baseline; being matched by a sub-10B-active model (Qwen3.6-35B-A3B at 24.6%) on this benchmark is the load-bearing comparative finding.

See also: Qwen, Alibaba, MOC - Open Source Models, MOC - Agentic Coding