the AI bench
VERIFIED JUNE 2026
All models

MODEL · ALIBABA · 480B TOTAL / 35B ACTIVE

Qwen3-Coder-480B-A35B

Frontier open-weight agentic coding model — claimed on par with Claude Sonnet on agentic benchmarks. Alibaba's most powerful coder.

License: Apache 2.0 · Context: 262K native, extendable to 1M · Released: July 2025

The decision in five lines

The call
Hosted only
Best for
Hosted reference and benchmarks
Runs on
Hosted or workstation-class only · ~240 GB (not consumer-local)
Watch out
Anything local under a workstation-grade multi-GPU rig — at Q4 it's still ~240 GB.
Evidence
Estimated · last verified April 2026

480B total
PARAMETERS
MOE
TYPE
262K
CONTEXT
~240 GB (not consumer-local)
VRAM AT Q4

Where we recommend this

This model isn’t currently in an active planner slot. See the runner notes below if you’re running it anyway.

The call

Frontier open-weight agentic coding model — claimed on par with Claude Sonnet on agentic benchmarks. Alibaba's most powerful coder.

When not to use: Anything local under a workstation-grade multi-GPU rig — at Q4 it's still ~240 GB. Practically a hosted/API or dual-server-class model.

Runner notes

MLX 4-bit build exists (`mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit`) for Mac Studio 512 GB. Otherwise FP8 on 8× H100 or hosted via Qwen Code / Cline. Function-call format is custom — use Qwen's template, not OpenAI's.

License
Apache 2.0
Released
July 2025
Maker
Alibaba

Next step

Find-by-model — see what hardware runs this