modelgrep
Q

Qwen: Qwen3.5-35B-A3B

qwen/qwen3.5-35b-a3b

78th smartest of 178Cheaper than 76% of paidReasoningToolsJSONVision
Use via OpenRouter ↗
Intelligence
30.7
78th of 178
Design Elo
Speed
180
17th fastest
Latency
333ms
first token
Input price
$0.140
73rd cheapest
Context
262K
82K max out

How it compares

Smarter than56%
of all ranked models
Faster than94%
of all ranked models
Cheaper than76%
of all ranked models

Overview

The Qwen3.5 Series 35B-A3B is a native vision-language model designed with a hybrid architecture that integrates linear attention mechanisms and a sparse mixture-of-experts model, achieving higher inference efficiency. Its overall...

Benchmarks

independent · via OpenRouter
Artificial Analysis54th percentile
Intelligence Index
30.7
Coding Index
16.8
Agentic Index
48.0
GPQA Diamond
82%
Humanity's Last Exam
13%
SciCode
29%
Tau²-Bench (agentic)
86%

Providers & pricing (9)

ProviderIn $/MOut $/MUptime
DeepInfrafp8$0.140$1.00
Parasailfp8$0.150$1.00100%
AkashMLfp8$0.160$1.20100%
Alibaba$0.163$1.30100%
AtlasCloudfp8$0.225$1.80
SiliconFlowfp8$0.240$1.80
WandBfp8$0.250$1.25
NextBitfp8$0.300$1.80
Venice$0.313$1.25

Specifications

Context window262K
Max output82K
Knowledge cutoff
Input modalitiestext, image, video
Output modalitiestext
Prompt caching
Cache read price$0.050/M
ModeratedNo

Qwen3.5-35B-A3B FAQ

How much does Qwen3.5-35B-A3B cost?

Qwen3.5-35B-A3B costs $0.140 per million input tokens and $1.00 per million output tokens via OpenRouter, making it 73rd cheapest of 298 paid models.

How smart is Qwen3.5-35B-A3B?

Qwen3.5-35B-A3B scores 30.7 on the Artificial Analysis Intelligence Index, ranking 78th of 178 benchmarked models, with a GPQA Diamond score of 82%.

How fast is Qwen3.5-35B-A3B?

Qwen3.5-35B-A3B generates around 180 tokens per second with 333ms time-to-first-token (p50), the 17th fastest tracked model.

What is Qwen3.5-35B-A3B's context window?

Qwen3.5-35B-A3B supports a 262K-token context window and can output up to 82K tokens. It accepts text, image, video input.

Compare head-to-head