asoning mix
text, image, and audio data
with all reasoning traces removed
Non-reasoning checkpoint
answers every verifiable prompt
eight times, thinking off
Verifiable prompts
OpenMathReasoning, DeepMath,
NuminaMath, HARP, IF single-turn
Verify the eight answers
pass@8
Add a reasoning trace
generated by thinking
models and verified
Production text mixture
trains 8B and 70B
≥1\geq 1 of 88 correct
00 of 88 correct
Figure 9: Capability filtering. We train an auxiliary nonthinking Apertus 1.5 8B checkpoint to evaluate verifiable text prompts. Prompts solved at least once in eight attempts enter the production text mixture without reasoning traces; those that fail all eight receive verified traces. Both production models use the resulting mixture.
Data.
The text component of the final production SFT mixture contains 3,707,851 samples (Table 6). Relative to Apertus 1.0, it shifts weight toward instruction following, mathematics, and code, adds newer multilingual chat data, carries refusal-style safety data whose responses we regenerate and verify as described below, and carries roughly 250,000 samples that reference the Apertus identity, mostly through their system prompts, though no dedicated identity question-answer pairs. The reasoning traces come partly from openly released datasets whose traces are distilled from open reasoning models, and partly from traces we generate ourselves on Alps. For mathematics and instruction following, we generate verified reasoning traces with thinking enabled, using multiple models: Kimi-K2.6, Qwen3.6-27B, Gemma-4-31B, and MiniMax-M2.7. We deduplicate text prompts, and for the capability-filtering phase we filter out prompts that could lead to verifiable answers, such as math prompts, using an LLM-as-a-judge (Kimi-K2.6). The text mixture remains thin on long multi-turn conversations, with only roughly 30,000 samples of three or more turns.
The training mixture for the auxiliary checkpoint that we subsequently use for capability filtering of text data includes an image pool of 20 datasets with roughly 58B tokens and an audio pool of six datasets with roughly 1.1B tokens (Table 7, Table C.4). For this checkpoint’s multimodal supervision, we follow a zero-reasoning policy: we regenerate samples whose sources contain reasoning traces with Qwen3.6-27B and thinking disabled. The final production SFT mixture uses the expanded image pool in Table B.5, including reasoning-bearing additions described in Section 4.2.1, and the same prepared audio pool (Table C.4).
Table 6: Text component of the final production SFT mixture by source and category.
All datasets are decontaminated against evaluation benchmarks. Numbers indicate
example count after filtering. Data Ratio is each category’s share of
the text SFT pool by example count.
Category
Dataset Source
Samples
Data Ratio
Foundation
OLMo2 OASST1 (Tulu-3) (Köpf et al., 2023)
7,028
6.14%
Llama-Nemotron Chat (Bercovich et al., 2025)
20,585
Llama-Nemotron Safety (Bercovich et al., 2025)
13,765
Apertus FAQ (Hernández-Cano et al., 2026)
1,580
IF-augmented (single- & multi-turn)
184,580
Subtotal
227,538
Math & Reasoning
OpenMathReasoning (Moshkov et al., 2025)
666,344
25.64%
DeepMath-103K (He et al., 2026)
101,950
NuminaMath 1.5 (Li et al., 2024a)
83,818
SciRIFF (Wadden et al., 2025)
66,322
Table-GPT (Li et al., 2024c)
13,222
FeTaQA (Nan et al., 2022)
10,330
RiddleSense (Lin et al., 2021)
4,531
HARP (Yue et al., 2024)
4,128
Subtotal
950,645
Code & Functions
OpenCodeReasoning v2 (Ahmad et al., 2025)
512,428
19.79%
Glaive Function Calling v2 (Glaive AI, 2023)
52,727
Jupyter Agent (Colle et al., 2025)
29,529
GSM8K (TIR) (Cobbe et al., 2021)
7,473
Toucan 1.5M
117,586
EnvScaler SFT Trajectories
9,022
OpenSeeker v1
4,949
Subtotal
733,714
Multilingual
SmolTalk2 Multilingual (8 lang) (Bakouch et al., 2025)
653,834
25.83%
EuroBlocks Multilingual (2512) (Ramos et al., 2026)
286,381
EuroBlocks Aya (Synthetic 1124) (Martins et al., 2025)
17,467
Subtotal
957,682
Regional
WikiQA 