Back to Explore
Qwen: Qwen3 Coder 30B A3B Instruct
Alibaba·Released 2025-07-31
Open Source30B160K ctxApache 2.0
About
Qwen3-Coder-30B-A3B-Instruct is a 30.5B parameter Mixture-of-Experts (MoE) model with 128 experts (8 active per forward pass), designed for advanced code generation, repository-scale understanding, and agentic tool use. Built on the...
Related Models
Pricing
Input
$0.07
per 1M tokens
Output
$0.27
per 1M tokens
Blended
$0.12
per 1M tokens
Cheaper than 82% of models. Median price is $0.54/1M tokens.
Cost Calculator
Tokens per day1M
100K100M
Daily
$0.12
Monthly
$3.60
vs. Similar Models
Qwen: Qwen3 Next 80B A3B InstructQ:+0.1
$0.34+185%
QwQ 32BQ:-0.2
$0.74+521%
Qwen3 235B A22B (Reasoning)Q:-0.2
$2.63+2087%
Qwen3 VL 30B A3B (Reasoning)Q:-0.3
$0.34+182%
Performance
111
tokens/sec
Faster than 59% of models
1.51
seconds
Faster than 33% of models
1.51
seconds
Faster than 56% of models
Market Median
94 tok/s
17% faster
Median TTFT
1.11s
36% slower
Throughput/Dollar
922
tok/s per $/1M
Speed Comparison
GPT-5.4 (Non-reasoning)
111 tok/s+0%
LFM2 24B A2B
111 tok/s+1%
Claude 4.5 Haiku (Reasoning)
111 tok/s+1%
Context Window
160K
tokens
Larger than 48% of models
Max Output
33K
tokens
20% of context
Benchmarks
MMLU-Pro
70.6%
GPQA Diamond
51.6%
HLE
4.0%
LiveCodeBench
40.3%
SciCode
27.8%
TerminalBench Hard
15.2%
MATH-500
89.3%
AIME
29.7%
AIME 2025
29.0%
IFBench
32.7%
Long Context Recall
29.0%
Tau2
34.5%
Market AverageTop Score
Open Source
apache-2.030BGGUF / GPTQ / AWQ
Downloads
1.9M
Likes
1.1K
VRAM (FP16)
24-48 GB
GPU
A6000 / M3 Ultra