ing and development splits and complete final packing, the pool of the Apertus 1.5 8B run contains 43,915,058 sequences and 143.4B input tokens, and Table B.5 gives its per-dataset composition.
Table 7: Image pool used to train the auxiliary checkpoint for capability filtering of text data: 20 datasets, 17.5M sequences, 58.1B tokens. The production SFT pool appears in Table B.5.
Dataset
Sequences
Tokens (M)
Processing
llava_ov_permissive
(An et al., 2025)
10,019,044
24,859.2
Permissive subset
sensenova_si_8m
(Cai et al., 2026)
5,205,020
21,859.4
Reasoning stripped
google_maptrace
(Panagopoulou et al., 2025)
1,055,208
7,585.8
As released
lnqa_recap
(Pont-Tuset et al., 2020; Wiedmann et al., 2025)
302,777
1,034.5
Answers regenerated
pixmo_cap_qa
(Deitke et al., 2025)
163,359
768.3
As released
CoSyn_400k
(Yang et al., 2025b)
87,057
359.1
As released
bigearthnet
(Herzog et al., 2026)
229,114
281.9
As released
google_rsrcc
(Kazoom et al., 2026)
103,867
233.4
Reasoning stripped
pixmo_ask_model_anything
(Deitke et al., 2025)
58,830
203.4
As released
mmevol_judged
(Luo et al., 2025)
90,216
197.7
Judge filtered
tcm_shizhen_vision
(Chen et al., 2025c)
63,806
185.6
As released
molmo2_syn_multiimage
(Clark et al., 2026b)
30,286
160.3
As released
molmopoint_guisyn
(Clark et al., 2026a)
20,939
97.5
As released
vdr_cooking_recipes
(racine.ai et al., 2025)
11,649
80.0
As released
chinesememe_recap_en
(LLM-Red-Team Contributors, 2024; Wiedmann et al., 2025)
42,678
73.2
Recaptioned
molmo2_multiimage_qa
(Clark et al., 2026b)
13,141
63.4
As released
path_vqa
(He et al., 2020)
19,654
32.3
As released
multihiertt
(Zhao et al., 2022)
5,358
22.9
As released
memotion_recap_en
(Ramamoorthy et al., 2022; Wiedmann et al., 2025)
5,857
10.0
Recaptioned
spatialsense_gold
(Yang et al., 2019)
6,812
6.4
Answers regenerated
Total
17,534,672
58,114.2
Categorical Answer Balance.
We inspect categorical answers in image SFT sources whose schemas permit deterministic extraction. Their distributions vary substantially by source. In RSRCC, 97.6% of 52,944 binary answers are “yes.” Among 50,923 questions with four options, 40.8% use answer A, 54.3% use B, 4.8% use C, and 0.1% use D. By contrast, the correct answers in BigEarthNet are almost uniformly distributed across the four options, and its binary answers are close to balanced; SpatialSense is exactly balanced between “yes” and “no” (Figure 10). The RSRCC distribution can induce answer priors and can inflate evaluation scores on data drawn from the same source.
Figure 10: Representative categorical answer distributions in image SFT sources with deterministic extraction. Left: RSRCC strongly favors the first two answer options among questions with four options, whereas BigEarthNet is nearly uniform. Right: RSRCC also strongly favors affirmative binary answers, unlike BigEarthNet, SpatialSense, and PathVQA. These source distributions can create answer priors but do not by themselves establish a causal effect on evaluation accuracy.
4.2.2 Audio
The audio pool used to train the auxiliary checkpoint for capability filtering of text data is speech only, with no music or environmental-sound supervision: six datasets with 1,116,511 sequences and roughly 1.1B tokens, all speech in and text out. We synthesize and filter these datasets as described in Section 2.2.4. The production SFT runs draw from the same six-dataset pool listed in Table C.4. These counts describe prepared data available to the runs, rather than sampling weights, repetition factors, or tokens actually consumed. The later long-context SFT stage draws from the extended pool in Table C.5; Table 8 gives its per-model training counts.
4.2.3 Variety-Aware Translation Case Study
Generating text in the correct variety of a multi-variety language can be a challenge for language models, especially when some varieties are underrepresented in the training data. Hernández-Cano et al. (2026) studied this problem for Romansh, a Swiss national language with six distinct v