Skip to main content
Meo

Qwen: Qwen2.5 VL 72B Instruct

Qwen: Qwen2.5 VL 72B Instruct is qwen's model. On the un-leaked meo benchmark it ranks #61 of 82 with 50.2% accuracy and an Effective Value (𝕍) of 0.0002 β€” earning a correct answer for about $0.0009 in 35s each.

as of 2026-10-18

meo rank
#61
of 82
Accuracy
50.2%
overall
𝕍 index (top=100)
0.005
raw 𝕍 0.0002 Β· N=10
$ / correct
$0.0009
35s each

Per-domain accuracy

127 correct of 250 questions asked across 11 domains Β· answer rate 100% Β· macro accuracy 48.8%. Headline accuracy is micro-averaged per meo’s methodology; per-domain question counts (correct/asked) are shown so you can judge significance.

Theory of mind
100%3/3
Tape machine
73%11/15
Base-rate bias
73%30/41
Regex automaton
69%11/16
Framework bias
67%2/3
Perceptual illusions
51%52/102
Watson-Glaser
50%2/4
Long arithmetic
27%4/15
CSP (unsatisfiable)
25%4/16
Logic / math / CS
24%8/34
State tracking
0%0/1

Efficiency

Cost per correct
$0.0009
Seconds per correct
35
Tokens per correct
933
Correct answers
127 / 250
Total benchmark cost
$0.12
Avg tokens / item
474

Model details

Released
2025-02-01
Weights
Open
Context
128K
Max output
115K
Knowledge cutoff
2024-06-30
Price in
$0.80/M
Price out
$1.00/M
Modality
text+image->text
View on OpenRouter β†—Hugging Face β†—

Sources: openrouter Β· meo

FAQ

How accurate is Qwen: Qwen2.5 VL 72B Instruct?

Qwen: Qwen2.5 VL 72B Instruct scores 50.2% overall on the meo benchmark (rank #61 of 82), an un-leaked private holdout graded objective-first.

What is Qwen: Qwen2.5 VL 72B Instruct's Effective Value?

Its Effective Value (𝕍) is 0.0002 at N=10, Ο‰=1, Ξ΄=1.5 β€” a fused metric of accuracy, speed, cost, and error-cascade. 𝕍 rankings shift with task depth.

How much does Qwen: Qwen2.5 VL 72B Instruct cost per correct answer?

About $0.0009 per correct answer (933 tokens, 35 seconds each) at benchmark-time pricing via OpenRouter.

Is Qwen: Qwen2.5 VL 72B Instruct open weights?

Yes β€” Qwen: Qwen2.5 VL 72B Instruct is released with open weights.

Benchmark updates affecting this model

  • 2026-06-09majorlogic_math_cs ground truth was broken β€” culled and rebuilt with solver-verified items

    A 5-lab frontier jury revealed that 29 of 37 LLM-authored logic_math_cs items had broken ground truth: on most, five independent frontier models unanimously concluded the clue sets were contradictory while the stored answer claimed a number, and on eight, all five agreed on the SAME different number (author arithmetic errors). Models were being marked wrong for being right. The 29 were retired (full audit published in the repo) and the domain rebuilt with generator-as-oracle items: an embedded constraint solver PROVES each puzzle has exactly one solution (or provably none), verified by an independent exhaustive enumerator. A 3-round difficulty study found frontier models saturate consistent human-scale logic puzzles, so items use anonymous global constraints and carry a measured difficulty gradient.

  • 2026-06-08majorCorrected base_bias grading (LLM-equivalence judge fix)

    The LLM-equivalence grader uses a reasoning model whose verdict token budget was too small, so its YES/NO answer was emitted as empty and silently counted as a failure. This understated the base_bias domain to ~7% for every model when the true accuracy is ~85-90%. We raised the judge budget, excluded its reasoning tokens from the verdict, switched to verbosity-robust conclusion matching, and re-graded the whole board.

Full benchmark changelog β†’

Embed this ranking

Show Qwen: Qwen2.5 VL 72B Instruct’s meo rank on your site β€” the badge links back here.

Related: all Qwen (Alibaba) models Β· Theory of mind domain Β· best Effective Value Β· biggest movers Β· methodology Β· full leaderboard.