sen responses, so we cannot exclude a self-preference effect.
I.4 Completion Generation
Every model answers every prompt once, at temperature 1.0 and a cap of 4,096 generated tokens. We leave top-pp and top-kk unset, so each model falls back to the defaults of its own shipped generation configuration; we therefore sample the pool at a common temperature but not under a single decoding configuration, and the Qwen3 models in particular carry non-neutral nucleus defaults. We disable thinking throughout, through the chat template for every model that accepts the argument and through a no-reasoning launch flag for the four Mistral models whose tokenizer rejects it.
I.5 Score Distribution
Max-min selection over a pool of very unequal capability pins both ends of the scale: 86.5% of chosen responses score in [4.9,5.0][4.9,5.0] and 96.7% of rejected ones sit at the 1.0 floor, giving a mean margin of 3.766 (Table I.3 and Table I.4). The margin is large almost everywhere, with 97.9% of pairs separated by at least two points and 93.4% by at least three. The 3,600-token filter costs little separation: the mean per-prompt max-min gap is 3.822 before filtering against a realized 3.766 after it.
Table I.3: Summary of the four-aspect judge scores over the 245,898 preference pairs: the chosen score, the rejected score, and their margin.
Mean
Std
p5
Median
p95
Chosen
4.897
0.354
4.345
4.998
5.000
Rejected
1.132
0.385
1.000
1.003
1.738
Margin
3.766
0.503
2.804
3.974
4.000
Table I.4: Preference pairs by score band.
Score Band
Chosen
%
Rejected
%
[1, 2)
527
0.21
238,016
96.73
[2, 3)
1,872
0.76
5,264
2.14
[3, 4)
5,425
2.21
1,586
0.64
[4, 4.5)
7,414
3.02
544
0.22
[4.5, 4.9)
17,868
7.27
297
0.12
[4.9, 5.0]
212,792
86.54
191
0.08
Appendix J Alignment Ablations
J.1 DPO Ablations
This appendix reports the sweeps over the learning rate, the KL penalty (β\beta), and the batch size for offline DPO (Rafailov et al., 2023), its length-normalized variant (Meng et al., 2024; Lambert et al., 2025), and online DPO (Guo et al., 2024). It also compares strategies for combining offline and online DPO. Section 4.6 documents the final configurations and Section 5.3.2 their downstream results.
All scores are percentages, and Δ\Delta is the difference in average from the SFT model that initializes the runs. The ablations use intermediate checkpoints and earlier versions of the evaluation suite; scores are therefore comparable only within a table. In particular, AlpacaEval uses Llama-3.3-70B-Instruct (Grattafiori et al., 2024) and Arena-Hard uses Qwen3.5-27B (Qwen Team, 2026b) as the judge, instead of the GPT judges of the post-training lineage (Section 5.3.2); these scores, and the averages that include them, are not comparable to Table 26. For Apertus 1.5 70B we report skill-group means.
J.1.1 Apertus 1.5 8B
Offline DPO, standard loss.
Table J.1 lists the ten best configurations of a 60-run grid over the learning rate, β\beta, and the batch size. Their averages lie within 0.5 points. Nine of the ten use a learning rate of 5⋅10−75{\cdot}10^{-7} or 10−610^{-6}, and none uses β≥0.5\beta\geq 0.5. All three batch sizes appear among them, which indicates a low sensitivity to the batch size. The best configuration (β=0.05\beta{=}0.05, learning rate 5⋅10−75{\cdot}10^{-7}, batch size 512) reaches Δ=+4.1\Delta{=}{+}4.1.
Table J.1: Offline standard DPO from Apertus 1.0 8B SFT (Hernández-Cano et al., 2026): the ten best configurations of a 60-run grid over learning rate ∈{10−6,5⋅10−7,10−7,5⋅10−8}\in\{10^{-6},5{\cdot}10^{-7},10^{-7},5{\cdot}10^{-8}\}, β∈{0.01,0.05,0.1,0.5,1.0}\beta\in\{0.01,0.05,0.1,0.5,1.0\}, and batch size ∈{128,256,512}\in\{128,256,512\}, sorted by average; the learning-rate row is in units of 10−710^{-7}. Scores are percentages (higher is better). The two rightmost columns are Apertus 1.0 8B SFT, from which all runs start, and Apertus 1.0 8B Instruct. Δ\Delta is the difference in average from the SFT column, and – marks the SFT column itself. Bold marks the be