Vision ArenaLeaderboard
Answer prompts that include an image; real users vote blind between two models' answers.
1,219
37% expected win rate against the board leader (1,313)
#44 of 54
board 2026-08-27
Source: ArenaAlibabaChina· released Feb 2026· open weights
Qwen3.5 27B
Provisional: 5 comparable tests so far. Ranked once it reaches 6 across 3 counting categories.
Alibaba, China. One of 19 models from this lab on the Index.
Third-party publishedLow confidence (thin coverage)
Snapshot September 8, 2026
OpenCharts Index
43.6
Provisional: not enough comparable tests or counting categories to rank yet · 5 comparable tests · 0 counting categories
Drop any single test and the Index lands between 33.6 and 58.1.
Needs more results (0 of 8 so far)
Distance to today's frontier on the hardest tests. Not a measure of intelligence.
What it's known for
Qwen3.5 27B, in 5 results.
Its widest gap is on Vending-Bench 2, 100.0% behind the best published result. No comparable results yet in reasoning, knowledge, math and long context. Between 50% and 60% behind the best published results, on average.
Strongest category
No category counts yet: a category counts once it has more than one comparable test and the model has taken at least half of them. The scores it has are below.
Best single result
Vision Arena
Answer prompts that include an image; real users vote blind between two models' answers.
1,219
37% expected win rate against the board leader (1,313) · #44 of 54.
Seeing is half of most real tasks: screenshots, documents, photos and charts.
Where it trails
Vending-Bench 2
Run a simulated vending business for a year: ordering, pricing and cash flow. The score is the final balance.
$202
100.0% behind the best published result ($11,182) on a log scale · #44 of 53.
Long-horizon decisions whose consequences compound. It exposes models that drift or forget.
Not measured yet
No comparable result yet in these 4 categories. A blank is a blank, never a zero, and it does not lower the Index.
- Reasoning
- Knowledge
- Math
- Long context
Covered, but not averaged into the Index:
- Coding2 of 7 tests · needs 4 · mean 28.7
- Agentic1 of 8 tests · needs 4 · mean 0.0
- Multimodalsingle-test category, shown beside the Index · 73.5
- Human preferencesingle-test category, shown beside the Index · 72.2
See it at work
What Qwen3.5 27B was asked to do, and how it did.
Every test it has taken, by category, with the kind of task it faced, who produced the number and where the result landed against the best published one. The examples are original and illustrative, never items from the datasets themselves.
Multimodal · mean 73.5 across 1 comparable test · a single-test category, shown beside the Index and never averaged into it
Every result is read against the best published one on its test: percent scores as a share of the best above chance, Arena ratings as the expected win rate against the board leader, open-ended values on a log scale. Sitting a harder exam never lowers a model. Every number links to the publisher that produced it.
Head to head
Qwen3.5 27B against whoever you pick.
Choose a rival. Every counted test both have taken appears side by side as a share of the best published result on that test, with the category means above. A win is a higher share; a gap under 2 points is a tie, because one item on a 45-problem exam is 2.2 points and publishers' own standard errors are of that order.
Qwen3.5 27B
provisional·Index 43.6D
0 : 4
wins · 0 ties · 4 shared
Claude Fable 5.1
#2·Index 97.0A+
Claude Fable 5.1 comes out ahead on 4 of the 4 tests both have taken.
Category means
Reasoning
—97.5Knowledge
—93.7Coding
28.797.0Math
—96.6Agentic
0.088.4Multimodal
73.5—Human preference
72.299.2
Shared tests · biggest gaps first
- Vending-Bench 2Agentic−76.7$202$5,422
- Code Arena (WebDev)Coding−75.31,3571,762
- WeirdMLCoding−57.439.5%92.9%
- Text ArenaHuman preference−27.01,4081,504
Bars are distances to the best published result on each test (percent scores above chance, Arena ratings as win rate against the leader, open-ended values on a log scale). Raw values as the publishers report them.
Open Claude Fable 5.1Scores
Every test, every source
The best published run per benchmark, the raw value as the publisher reports it, who produced the number, how close it comes to the best published result on that test, and where the model sits among every model scored on it. Percent scores are read above chance, Arena ratings as a win rate against the board leader, open-ended values on a log scale. A test with too few models to compare against is shown but not counted.
Category means
- Reasoning—
- Knowledge—
- Coding28.7 · 2/7
- Math—
- Agentic0.0 · 1/8
- Multimodalbeside73.5 · 1/1
- Human preferencebeside72.2 · 1/1
- Long context—
Mean · tests taken of the category's comparable tests. A faded category either rests on fewer than half of them or is a single-test category (marked “beside”), which is shown next to the Index and never averaged into it.
WeirdMLLeaderboardCoding
39.5%
42.6% of the best (92.9%)
10th percentile of 69
Code Arena (WebDev)LeaderboardCoding
1,357
7% expected win rate against the board leader (1,797)
19th percentile of 86
board 2026-09-05
Source: Arena (LMArena) — Leaderboard DatasetVending-Bench 2LeaderboardAgentic
$202
0.0% of the best ($11,182) on a log scale
0th percentile of 53
Vision ArenaLeaderboardMultimodal
1,219
37% expected win rate against the board leader (1,313)
19th percentile of 54
board 2026-08-27
Source: Arena (LMArena) — Leaderboard DatasetText ArenaLeaderboardHuman preference
1,408
36% expected win rate against the board leader (1,507)
12th percentile of 92
board 2026-09-02
Source: Arena (LMArena) — Leaderboard DatasetTop of the leaderboard
Compare with its neighbours.
The best models, ranked here, working inside Theo.
28 of the models on this page run inside Theo today. Theo picks the right one for each step and always shows which engine answered.