Can a vision-language model look at a picture of a chain reaction and tell what will happen? Recent multimodal models describe scenes fluently and solve textbook physics problems, which suggests they should also predict how a scene of ramps, ledges, pendulums, seesaws and dominoes plays out once a ball is pushed. We tested this claim on 30 models from 10 families and reached a clear conclusion: unaided, no model predicts the outcome reliably. We built DominoBench, a benchmark of still images of 2D rigid-body scenes whose outcome a physics simulator computes. It holds 854 items in 427 minimal pairs, in which a change of one to three grid steps in one quantity decides whether one person survives, and every item comes with a four-rung hint ladder ending in the exact calculation that decides the outcome. Unaided, frontier models from Anthropic, Google and OpenAI are right about the person whose fate flips in at most 75% of items and answer both twins of a pair correctly in at most 40%. Pointing at the deciding step helps little, and exact geometry given as text barely helps, yet once the deciding calculation and the quantities it needs, including speeds that a still image cannot show, are written out, the same models reach 80-92% on the deciding person, while small open-weight models stay near chance. Physical prediction from a still image therefore fails at the quantitative step: the strongest models use the deciding quantity when they are given it, but do not produce it on their own. For models that plan or act in the physical world, a fluent account of a scene is thus no evidence of a correct prediction: such models must find and compute the deciding quantity unprompted, and DominoBench tests whether they do.
Figures & tables
Figure 1: A minimal pair from DominoBench-1K, also in DominoBench-124. Left: the still of item (a). A pushed stone ball sets off two rows of dominoes, and a small ball drops into the bucket of a spiked seesaw above C. Item (b) differs only in a counterweight one grid step (0.25 squares) wider. Zooms: in (a) the ball outweighs the counterweight and the spikes hit C; in (b) the seesaw does not tip. Right: the question and hints, with the H3 value for (b) in parentheses. Each rung is a fresh query with all hints up to it. Unaided, Claude Opus 5.5 gets both items wrong, swapping C’s fates; with H3, it gets both right. Data: pair 4 (items 0004-a and 0004-b).
Exact
Both twins
Deciding person
DominoScore
Model
n
H0
H3
H0
H3
H0
H3
H0
H3
DominoBench-1K, full: all 854 items (427 pairs)
GPT-5.6-Sol
854
42
63
13
49
59
88
11
67
Claude Opus 5.5
854
47
62
16
44
61
85
12
54
Claude Opus 4.1
854
45
63
16
45
60
85
12
52
Claude Sonnet 5.5
854
42
64
15
49
57
86
12
50
Table 1: Accuracy (%) in image mode, unaided (H0) and with the calculation (H3), on DominoBench-1K and DominoBench-124 (DominoBench-38: supplement). n : items answered, complete runs only. Both twins : both items of a pair exactly right. Deciding person : the person whose fate flips between the twins. † Open weights. Baselines on DominoBench-1K: the naive chain reaction (all but the pusher die) and “everyone survives” score 14% and 25% exact, 0% on both twins and 50% on the deciding person, an oracle 100%. Bold: best per column and benchmark; ranked by the shaded column. Data: DominoBench-1K and DominoBench-124.
Figure 2: DominoScore of every model with a complete run on DominoBench-124, with the deciding calculation (H3, bars with ± 1 bootstrap standard error over twin pairs) and unaided (H0, open circles). Neighbouring ranks are often within error. Scores on DominoBench-1K are in Table 1 ; the small DominoBench-38 is in the appendix.
Figure 3: Accuracy on the deciding person along the hint ladder: H0 no hint, H1 a generic nudge, H2 a pointer to the deciding step, H3 the deciding calculation. Left: Claude. Right: open-weight models with a complete ladder. Solid: image mode; dashed: text only. Dotted line: 50%. Bars: 95% Wilson intervals. Data: DominoBench-124.
Figure 4: Accuracy on the deciding person by deciding mechanism, from H0 (open) to H3 (filled), image mode, all 854 items. Left: closed models, right: open-weight models above DominoScore 20 at H3. Data: DominoBench-1K.
Figure 5: Stated probability of death for the deciding person against how often that person dies, in bins of 0.2 (at least 15 statements each), image mode. Solid: closed models that answered every item; dashed: the five larger open-weight models of Fig. 4 , pooled; dotted: perfect calibration. Data: DominoBench-1K.
Appendix figures & tables14 assets
Supplementary material from the paper’s appendix.
Appendix
Table S2 : Full leaderboard on DominoBench-1K (all 854 items, 427 pairs), image mode, unaided (H0) and with the calculation (H3). Columns as in Table 1 . † Open weights. Bold: best in each column. Ranked by DominoScore at H3. Data: DominoBench-1K.
Table S3 : Full leaderboard on DominoBench-124 (subset of DominoBench-1K, 62 pairs), image mode, H0 and H3. Columns as in Table 1 . † Open weights. Bold: best in each column. Ranked by DominoScore at H3. Data: DominoBench-124.
Table S4 : Full leaderboard on DominoBench-38 (19 pairs, six mechanisms), image mode, H0 and H3. Columns as in Table 1 . DominoScore uses the frozen weights renormalized over the six mechanisms. † Open weights. Bold: best in each column. Ranked by DominoScore at H3. Data: DominoBench-38.
Figure S6 : DominoScore of every model on DominoBench-38, the one benchmark every model answered, with the deciding calculation (H3, bars with ± 1 bootstrap standard error over twin pairs) and unaided (H0, open circles). Only complete runs are shown. Scores rest on six mechanisms with the frozen weights renormalized, so neighbouring ranks are often within error. Figure 2 gives the same ranking on DominoBench-124.
Exact
Deciding person
Model
Mode
n
H0
H1
H2
H3
H0
H1
H2
H3
Unparsed
Claude Opus 5.5
image
124
43
44
47
61
59
60
65
91
0
Claude Opus 5.5
text
124
46
40
47
64
64
60
68
86
0
Claude Sonnet 5.5
image
124
41
40
52
60
58
60
73
86
0
Claude Sonnet 5.5
text
124
40
45
48
58
60
66
70
86
0
Claude Haiku 5.5
image
124
36
35
39
52
57
59
59
82
0
Appendix
Table S5 : DominoBench-124, the subset of DominoBench-1K (62 pairs, 124 items): accuracy (%) at each hint rung, by input mode. n : items answered in that mode. ∗ Run in progress, H0 and H3 only. Unparsed: the largest share (%) of unparsed answers across the row’s rungs. A dash: not run, or still running. † Open weights. Bold: best in each column (baselines and runs in progress excluded). Data: DominoBench-124.
Exact
Deciding person
Model
Mode
n
H0
H1
H2
H3
H0
H1
H2
H3
Unparsed
Claude Opus 5.5
image
38
45
50
42
58
66
68
66
90
0
Claude Opus 5.5
text
38
40
45
47
63
58
60
68
92
0
Claude Sonnet 5.5
image
38
42
47
58
55
58
66
82
95
0
Claude Haiku 5.5
image
38
34
29
37
53
63
53
60
95
0
Claude Opus 4.1
image
38
42
40
47
60
66
63
71
95
0
Appendix
Table S6 : DominoBench-38 (19 pairs, 38 items): accuracy (%) at each hint rung, by input mode. n : items answered in that mode. Unparsed: the largest share (%) of unparsed answers across the row’s rungs. † Open weights. Bold: best in each column (baselines and runs in progress excluded). Data: DominoBench-38.
Figure S7 : Share of twin pairs answered identically for both twins, unaided (H0, open) and with the deciding calculation (H3, filled), image mode, every model with a complete run. Colour: provider; squares and † : open weights. Data: DominoBench-124.
Ledge
Pit
Pendulum
Recess
Seesaw
Dominoes
Jump
Model
Items
H0
H3
H0
H3
H0
H3
H0
H3
H0
H3
H0
H3
H0
H3
Claude Opus 5.5
854
82
98
70
93
51
68
57
94
61
99
49
78
70
72
Claude Sonnet 5.5
854
70
91
62
93
52
61
49
94
55
100
54
95
68
71
Claude Haiku 5.5
854
68
68
62
84
57
66
51
94
55
98
55
80
71
71
Claude Opus 4.1
854
70
100
65
96
51
66
54
94
61
98
50
77
76
71
Claude Sonnet 4.5
124
43
79
25
60
43
64
60
95
44
100
60
95
56
61
Appendix
Table S7 : Deciding-person accuracy (%) by deciding mechanism, image mode, unaided (H0) and with the deciding calculation (H3). Items: all 854 items of DominoBench-1K, or the 124 items of DominoBench-124. † Open weights. Bold: best in each column; rows with 124 items rest on 14 to 20 items per cell and are not comparable with the 854-item rows. Data: DominoBench-1K and DominoBench-124.
H0
H3
Model
all
visible
all
visible
Gemini 3.8 Flash
3.1
3.2
62.5
62.9
Gemini 4 Hydrogen
26.6
30.6
61.1
65.9
Gemini 4 Argon
39.9
40.0
60.4
63.2
Claude Opus 5.5
8.6
8.5
59.9
63.8
Gemini 4 Carbon
40.8
42.4
57.3
59.7
Appendix
Table S8 : DominoScore on the medium benchmark before the removal of the pairs whose labelled death is not visible (all: 140 items) and after it (visible: DominoBench-124), image mode, H0 and H3. Data: DominoBench-124 and the 8 removed pairs.
Table S9 : Composition of DominoBench-1K and DominoBench-38. Counts are of items unless the row says pairs. A pair’s twin separation is the number of grid steps between the two values of its varied parameter (0.25 squares for lengths, 0.1 for ball radii). Data: DominoBench-1K and DominoBench-38.
Figure S8 : One minimal pair per deciding mechanism. Left: the still of item (a), with the person whose fate flips boxed. Right: that person in item (a) and in item (b), with the simulated fate. The row title names the parameter that differs and its value in each item (in grid squares). Data: DominoBench-1K.
Figure S9 : Exact prompts for item 0004-a (the pair of DominoBench: Evaluating Multi-Step Physical Reasoning from Still Images ). In image mode the picture is attached to the user prompt; in text mode the exact geometry is a JSON object of several thousand characters, shortened here.
Mechanism
Difficulty dm
Weight wm
Pendulum
0.345
0.300
Jump
0.287
0.250
Dominoes
0.175
0.152
Ledge
0.106
0.092
Pit
0.085
0.073
Recess
0.059
0.066
Appendix
Table S10 : Frozen DominoScore weights. Difficulty: error rate on the deciding person at H3 of the reference panel. Weight: normalized difficulty, floored at half the uniform weight and then renormalized. Data: DominoBench-1K.
Random guesser
Uniform weights
Benchmark
mean
95th pct.
models
Spearman ρ
DominoBench-1K
2.6
6.2
16
0.95
DominoBench-124
5.6
14.4
30
0.96
DominoBench-38
8.0
25.9
30
0.97
Appendix
Table S11 : DominoScore of a random guesser (mean and 95th percentile over 20,000 simulations) and the Spearman rank correlation at H3 between DominoScore with the frozen weights and with uniform weights, over every complete run. Data: all three benchmarks.