TAU2
Measured May 14, 2026Source
Score
0.44
Models15 2 
Qwen3 Coder 480B A35B Instruct
阿里巴巴阿里巴巴China
Qwen3 Coder 480B A35B Instruct is a large-scale, code-specialized language model from Alibaba's Qwen series. It features a Mixture-of-Experts (MoE) architecture with 480 billion total parameters and 35 billion active parameters, designed for high-performance code generation, understanding, and instruction following.
CodingReasoning
Input / 1M tokens
$0.30Output / 1M tokens
$1.80Output tokens/s
68.64First-token seconds
1.66sSupported plans
6Benchmark history
Evaluations
Terminalbench Hard
Measured May 14, 2026Source
Score
0.19
Lcr
Measured May 14, 2026Source
Score
0.42
Ifbench
Measured May 14, 2026Source
Score
0.4
Aime 25
Measured May 14, 2026Source
Score
0.39
Aime
Measured May 14, 2026Source
Score
0.48
Math 500
Measured May 14, 2026Source
Score
0.94
Scicode
Measured May 14, 2026Source
Score
0.36
Livecodebench
Measured May 14, 2026Source
Score
0.59
Hle
Measured May 14, 2026Source
Score
0.04
Gpqa
Measured May 14, 2026Source
Score
0.62
Mmlu Pro
Measured May 14, 2026Source
Score
0.79
Artificial Analysis Math Index
Measured May 14, 2026Source
Score
39.3
Artificial Analysis Coding Index
Measured May 14, 2026Source
Score
24.6
Artificial Analysis Intelligence Index
Measured May 14, 2026Source
Score
24.8
Plan availability
Products and plans that support this model
io.net Plan
io.net Plan
io.net is a decentralized GPU cloud platform providing subscription plans and pay-as-you-go access to a variety of AI models for chat, coding, content creation, and automation workflows.
Synthetic
Synthetic
Synthetic is a platform that runs open-source AI models in private, secure datacenters. It offers subscription and usage-based pricing for accessing models like DeepSeek, Llama, MiniMax, Kimi, and others, with OpenAI-compatible API access for use in coding agents and tools.
Discussion
Thinking... Make sure you are connected to GitHub server