modelgrep
Q

Qwen: Qwen3 VL 235B A22B Instruct

qwen/qwen3-vl-235b-a22b-instruct

135th smartest of 178ToolsJSONVision
Use via OpenRouter ↗
Intelligence
17.0
135th of 178
Design Elo
Speed
36
216th fastest
Latency
618ms
first token
Input price
$0.200
98th cheapest
Context
262K
16K max out

How it compares

Smarter than24%
of all ranked models
Faster than27%
of all ranked models
Cheaper than67%
of all ranked models

Overview

Qwen3-VL-235B-A22B Instruct is an open-weight multimodal model that unifies strong text generation with visual understanding across images and video. The Instruct model targets general vision-language use (VQA, document parsing, chart/table...

Benchmarks

independent · via OpenRouter
Artificial Analysis25th percentile
Intelligence Index
17.0
Coding Index
14.0
Agentic Index
19.2
GPQA Diamond
61%
Humanity's Last Exam
5%
SciCode
30%
Tau²-Bench (agentic)
27%

Providers & pricing (6)

ProviderIn $/MOut $/MUptime
DeepInfrafp8$0.200$0.88099.9%
Parasailfp8$0.210$1.9098.4%
Venicefp8$0.250$1.5091.9%
Alibaba$0.260$1.0499.9%
AtlasCloudfp8$0.300$1.5098.3%
Novitabf16$0.300$1.5096%

Specifications

Context window262K
Max output16K
Knowledge cutoffMar 2025
Input modalitiestext, image
Output modalitiestext
Prompt caching
Cache read price$0.110/M
ModeratedNo

Qwen3 VL 235B A22B Instruct FAQ

How much does Qwen3 VL 235B A22B Instruct cost?

Qwen3 VL 235B A22B Instruct costs $0.200 per million input tokens and $0.880 per million output tokens via OpenRouter, making it 98th cheapest of 298 paid models.

How smart is Qwen3 VL 235B A22B Instruct?

Qwen3 VL 235B A22B Instruct scores 17.0 on the Artificial Analysis Intelligence Index, ranking 135th of 178 benchmarked models, with a GPQA Diamond score of 61%.

How fast is Qwen3 VL 235B A22B Instruct?

Qwen3 VL 235B A22B Instruct generates around 36 tokens per second with 618ms time-to-first-token (p50), the 216th fastest tracked model.

What is Qwen3 VL 235B A22B Instruct's context window?

Qwen3 VL 235B A22B Instruct supports a 262K-token context window and can output up to 16K tokens. It accepts text, image input.

Compare head-to-head