AlibabaAlibaba·💬 Text Generation

Qwen 3 Coder 480B Turbo

CodeFunction CallingWeb Searchfp8private
🧠 Try in Intelligence →Try on Venice.ai ↗
Quick reference
Qwen 3 Coder 480B Turbo — TLDR
  • 🧠 Mixture-of-Experts coder: 480B total, 35B active parameters
  • ⚡ Turbo variant optimized for faster code-task inference^catalog^
  • 📏 256K native context, extendable toward 1M via extrapolation
  • 🔧 Built for agentic coding with function calling
  • 👁️ Non-thinking only — no reasoning-trace blocks generated
  • 🔒 Served here in FP8 quantization
  • 🌐 Integrates with Qwen Code, CLINE, and web search
  • 🏢 Developed by Alibaba's Qwen team
💰 Pricing
$0.350 / $1.50
per 1M · input / output
📏 Context
256K tokens
📅 On Venice since
Jan 27, 2026
127 days ago
Provider

Alibaba Group is a Chinese multinational technology company founded in 1999 and headquartered in Hangzhou, Zhejiang. Originally built around e-commerce and cloud computing, Alibaba has become one of the most prolific contributors to open-weight AI research,…

Read full profile →
46 models on Venice
17 text · 16 video · 5 image · 4 inpaint · 2 embedding · 2 tts
Since Jan 11, 2025

About this model

Qwen 3 Coder 480B Turbo is a latency-optimized serving variant of Alibaba's flagship Qwen3-Coder-480B-A35B-Instruct, a Mixture-of-Experts model with 480 billion total parameters and 35 billion active per inference. The "Turbo" label reflects inference-side optimizations for code workloads rather than a new architecture; here it runs in FP8 quantization. Like the base model, it natively supports a 256K-token context that can be extended toward 1M tokens using extrapolation techniques.

The model targets agentic coding: it is trained with a specially designed function-call format and integrates with tools such as Qwen Code and CLINE, plus the catalog's web-search capability. Notably, it operates only in non-thinking mode and does not emit separate reasoning-trace blocks, trading explicit step-by-step reasoning for direct, faster code generation.

Within the broader Qwen lineup on this catalog, it sits alongside general-purpose siblings like Qwen 3 235B A22B Instruct 2507 and the newer-generation Qwen 3.5 397B, but is purpose-built for code generation, tool use, and long-context reasoning over repositories. Per the catalog's own description, the Turbo build is optimized for faster inference on code tasks while preserving the underlying 480B MoE's capabilities. Because this is the newest entry in its Turbo coder family, no in-family predecessor is available for direct generational comparison.

This About section is AI-generated from public sources (Claude Opus 4.8), with no human editing. It may contain inaccuracies — verify critical details against the sources listed above.

Data sources: Venice API · HuggingFace · Wikipedia — enrichment updated 1d ago