Back to Explore
Qwen: Qwen3.5-122B-A10B
Alibaba·Released 2026-02-25
Open Source122B262K ctxApache 2.0Multimodal
About
The Qwen3.5 122B-A10B native vision-language model is built on a hybrid architecture that integrates a linear attention mechanism with a sparse mixture-of-experts model, achieving higher inference efficiency. In terms of...
Related Models
Pricing
Input
$0.26
per 1M tokens
Output
$2.08
per 1M tokens
Blended
$0.72
per 1M tokens
Cheaper than 45% of models. Median price is $0.54/1M tokens.
Cost Calculator
Tokens per day1M
100K100M
Daily
$0.72
Monthly
$21.45
vs. Similar Models
GLM-5 (Non-reasoning)Q:+0.1
$1.55+117%
OpenAI: o3 ProQ:+0.2
$35.00+4795%
Qwen3.5 397B A17B (Non-reasoning)Q:-0.3
$1.35+89%
MoonshotAI: Kimi K2 ThinkingQ:+0.4
$1.07+50%
Performance
145
tokens/sec
Faster than 71% of models
1.13
seconds
Faster than 49% of models
14.96
seconds
Faster than 32% of models
Market Median
94 tok/s
53% faster
Median TTFT
1.11s
1% slower
Throughput/Dollar
202
tok/s per $/1M
Speed Comparison
NVIDIA Nemotron Nano 9B V2 (Non-reasoning)
145 tok/s+0%
Grok 4.3 (medium)
144 tok/s-0%
OpenAI: o3
145 tok/s+1%
Context Window
262K
tokens
Larger than 62% of models
Max Output
262K
tokens
100% of context
Benchmarks
MMLU-ProNot evaluated
GPQA Diamond
85.7%
HLE
23.4%
LiveCodeBenchNot evaluated
SciCode
42.0%
TerminalBench Hard
31.1%
MATH-500Not evaluated
AIMENot evaluated
AIME 2025Not evaluated
IFBench
75.7%
Long Context Recall
66.7%
Tau2
93.6%
Market AverageTop Score
Open Source
apache-2.0122BGGUF / GPTQ / AWQ
Downloads
779.3K
Likes
578
VRAM (FP16)
Multi-GPU
GPU
8x A100 / H100