Skip to content
VECTOR WIREAI INTELLIGENCE
⌘K
UTC
Models741Deals30d269Policy30d65, 14 pendingSources1,014

Qwen3 VL 235B A22B Instruct vs Qwen3 VL 32B Instruct

Alibaba · released

Wins 6 of 7 areas

Coding · Agents · Reasoning · Facts · Images and charts · Following instructions

Alibaba · released

Wins 1 of 7 areas

Long documents

Qwen3 VL 235B A22B Instruct is the stronger all-rounder.Qwen3 VL 32B Instruct is cheaper and better at long documents.

Scores updated · 48 tests both models report · How we compare

Where each one wins

Tests won in each of the seven areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.

Long documents and following instructions rest on a single test each.

What it costs

Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.

Qwen3 VL 235B A22B Instruct$0.40 to read · $1.60 to write$2.00Price from Alibaba's own price page
Qwen3 VL 32B Instruct$0.16 to read · $0.64 to write$0.80Price from Alibaba's own price page

Qwen3 VL 32B Instruct costs 60% less for the same work.

The biggest differences

The tests each model wins by the widest margin, up to three each. Scores are out of 100.

Where Qwen3 VL 235B A22B Instruct pulls ahead

  • Completes tasks by operating a computer desktopOSWorld-Verified+34.1points aheadQwen3 VL 235B A22B Instruct66.7Qwen3 VL 32B Instruct32.6
  • Recent programming contest problemsLiveCodeBench v6+10.5points aheadQwen3 VL 235B A22B Instruct54.3Qwen3 VL 32B Instruct43.8
  • Code for real scientific research problemsSciCode+5.8points aheadQwen3 VL 235B A22B Instruct35.9Qwen3 VL 32B Instruct30.1

Where Qwen3 VL 32B Instruct pulls ahead

  • Reasons across sets of long documentsAA-LCR+1.6points aheadQwen3 VL 32B Instruct34.3Qwen3 VL 235B A22B Instruct32.7
  • Hard command-line tasks in a real terminalTerminal-Bench Hard+1.5points aheadQwen3 VL 32B Instruct8.3Qwen3 VL 235B A22B Instruct6.8

Every test, side by side

All 48 tests both models report. The winning score is in its model's colour; marks a score checked independently.

CodingQwen3 VL 235B A22B Instruct

Full coding ranking

AgentsQwen3 VL 235B A22B Instruct
  • OSWorld-VerifiedQwen3 VL 235B A22B Instruct by 34.166.732.6+34.1
  • GDPValQwen3 VL 235B A22B Instruct by 6.76.70+6.7

Full agents ranking

ReasoningQwen3 VL 235B A22B Instruct

Full reasoning ranking

FactsQwen3 VL 235B A22B Instruct

Full facts ranking

Images and chartsQwen3 VL 235B A22B Instruct
  • BLINKQwen3 VL 235B A22B Instruct by 3.470.767.3+3.4
  • MMMU-ProQwen3 VL 235B A22B Instruct by 3.367.664.3+3.3
  • OCRBenchQwen3 VL 235B A22B Instruct by 2.59289.5+2.5
  • MathVistaQwen3 VL 235B A22B Instruct by 1.184.983.8+1.1
  • CharXiv (RQ)tie62.162.8tie
  • MMStartie78.477.7tie

Full images and charts ranking

Long documentsQwen3 VL 32B Instruct
  • AA-LCRQwen3 VL 32B Instruct by 1.632.734.3+1.6

Full long documents ranking

Following instructionsQwen3 VL 235B A22B Instruct
  • IFBenchQwen3 VL 235B A22B Instruct by 3.542.739.2+3.5

Full following instructions ranking

Other results30 tests, not counted

Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.

Questions people ask

Which is better, Qwen3 VL 235B A22B Instruct or Qwen3 VL 32B Instruct?

Qwen3 VL 235B A22B Instruct wins six of the seven areas where both have results: coding, agents, reasoning, facts, images and charts and following instructions. Qwen3 VL 32B Instruct wins long documents, and costs 60% less.

Which is better for coding?

Qwen3 VL 235B A22B Instruct. It wins 2 of the 3 coding tests both models report; Qwen3 VL 32B Instruct wins 1.

Which is cheaper?

Qwen3 VL 235B A22B Instruct costs $0.40 per million input tokens and $1.60 per million output tokens; Qwen3 VL 32B Instruct costs $0.16 and $0.64. That makes Qwen3 VL 32B Instruct about 60% cheaper for the same work.

How do you compare the two?

We use the 48 benchmark tests both models have published scores on. The verdict counts the 18 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 30 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.