What is the difference between Qwen3 VL 235B A22B Thinking and Hy4 preview?

Qwen3 VL 235B A22B Thinking (Alibaba Cloud) features 235B with a 131k context window, while Hy4 preview (Tencent) features Open Weights with a 1024k context window.

Which is cheaper for production inference, Qwen3 VL 235B A22B Thinking or Hy4 preview?

Qwen3 VL 235B A22B Thinking input pricing is Open / Free per 1M tokens vs Hy4 preview at Open / Free per 1M tokens.

Can I run Qwen3 VL 235B A22B Thinking or Hy4 preview locally on private GPU hardware?

Qwen3 VL 235B A22B Thinking is an open-weights model and can be hosted locally using Ollama or vLLM. Hy4 preview is an open-weights model and can be hosted locally.

Back to LLM Benchmark & Hardware Sizing Engine

Alibaba CloudvsTencent

Qwen3 VL 235B A22B Thinking vs Hy4 preview

Side-by-side technical showdown between Qwen3 VL 235B A22B Thinking and Hy4 preview on TheModelverse. Compare verified LLM benchmark scores, quantization compression, local GPU hardware sizing, and API inference pricing.

Executive Showdown & Winner Breakdown

Who Wins Where: Qwen3 VL 235B A22B Thinking vs Hy4 preview

Verified across benchmarks, pricing & local VRAM footprint

Context Window Capacity

Context Depth

Leader:Hy4 preview(131k vs 1024k)

Accommodates larger single-turn document ingestions and extensive conversation histories.

Hardware Sizing & Model Compression

LLM Hardware Sizing & Quantization Sizing

Simulate weight compression levels and dynamic KV-cache expansion to verify whether these models fit on your local hardware or cloud GPU cluster.

Quantization Level

Active Precision

4 bits / parameter

VRAM Reduction

-72.5% vs FP16

Benchmark Retention

96–98% (Sweet Spot)

Context Simulator

Total VRAM Footprint @ 8k Context (INT4 (GGUF / AWQ))

Model 1Open Weights

Qwen3 VL 235B A22B Thinking

150.3GB

Weights: ~129.3 GBKV Cache: ~1.4 GB+15% CUDA Buffer

Total Params:

235B

Active Compute:

Dense

Attention Scheme:

GQA (Grouped-Query)

Max Context:

131k tokens

Target Hardware: 4x–8x H100 Cluster with Tensor Parallelism (TP=4)

Model 2Cloud Hosted

Hy4 preview

Zero Local VRAM

Inference runs fully on provider cloud infrastructure. Zero GPU required locally.

Total Params:

Open Weights

Active Compute:

Dense

Attention Scheme:

GQA / Multi-Head

Max Context:

1024k tokens

Target Hardware: Zero Local VRAM (Managed Cloud API)

GPU & Hardware Tier Compatibility Matrix

Hardware Configuration	Available VRAM	Qwen3 VL 235B A22B Thinking	Hy4 preview
16 GB VRAM RTX 4070 / 4080 (16GB), Mac M-Series (16GB)	16 GB	✕ OOM	Cloud API
24 GB VRAM 1x RTX 3090 / 4090 (24GB), Mac M-Series (32GB)	24 GB	✕ OOM	Cloud API
48 GB VRAM 2x RTX 4090 (TP=2), 1x L40S, Mac M-Series (64GB)	48 GB	✕ OOM	Cloud API
80 GB VRAM 1x NVIDIA A100 / H100 (80GB), Mac Studio (128GB)	80 GB	✕ OOM	Cloud API
160 GB Node 2x H100 (TP=2), 4x L40S, Mac Studio (192GB)	160 GB	Tight (Low Ctx)	Cloud API
Multi-Node Cluster 4x–8x H100 Datacenter Cluster	320 GB	Optimal	Cloud API

Inference Economics & Cost Simulator

Token Pricing & Scale Economics

Calculate estimated monthly cloud API spend and evaluate the breakeven point vs self-hosted GPU hardware.

Monthly Queries: 100,000

10k500k1M

Avg Prompt Tokens: 1000

Avg Output Tokens: 300

Pricing Metric	Qwen3 VL 235B A22B Thinking	Hy4 preview
Input Cost (/1M tokens)	Free / Open	Free / Open
Cached Input (/1M tokens)	—	—
Output Cost (/1M tokens)	Free / Open	Free / Open
Simulated Monthly Bill (100,000 calls)	$0 API Cost Open Weights	$0 API Cost Open Weights
Self-Hosted Breakeven vs $864/mo GPU	Self-Hostable Day 1	Self-Hostable Day 1

Architecture Matrix

Feature	Qwen3 VL 235B A22B Thinking	Hy4 preview
Release Date	9/23/2025	8/28/2026
Routing / MoE	Dense	Dense
Attention Mechanism	Standard / GQA	Standard / GQA
Source Type	Open Weights (Open)	Open Weights (Open)

Customize or Add a 3rd Model to this Showdown