SAGE Celer 2.6 Technical Card

SAGEA Research Team; Basab Jha; Firoj Paudel; Ujjwal Puri; Adrian Liu; Ethan Henkel; Zhang Yuting; Mateusz Kowalczyk

cs.CLMar 24, 2026

SAGE Celer 2.6 Technical Card

Authors: SAGEA Research Team, Basab Jha, Firoj Paudel, Ujjwal Puri, Adrian Liu, Ethan Henkel, Zhang Yuting, Mateusz Kowalczyk,

Organizations: SAGEA · Tribhuwan University | Vedas College · Tribhuwan University | Madan Bhandari Memorial College · Fudan University ·

Abstract

We introduce SAGE Celer 2.6, the latest in our line of general-purpose Celer models from SAGEA. Celer 2.6 is available in 5B, 10B, and 27B parameter sizes and benefits from extensive architectural modifications and further pre-training on an undisclosed model. Using our Inverse Reasoning (IR) pipeline, SAGEA natively trains Celer 2.6 to validate its own logic paths, minimizing cascading error and hallucination in complex reasoning tasks. Celer 2.6 also boasts natively integrated multimodal functionality with an end-to-end vision encoder to avoid common pitfalls in adapter-based approaches. Celer 2.6 provides highly competitive results on mathematics, coding, and general intelligence benchmarks (ACUMEN), along with low latency. Most importantly, Celer 2.6 is specifically optimized for South Asian language support, with a custom tokenizer for the Devanagari script and strong performance in both Nepali and Hindi without sacrificing English reasoning ability.

Figures & tables

Panel of Figure 1 — Figure 1: Performance overview of SAGE Celer 2.6 High relative to SAGE Actus 2.4 and Llama 3.3 70B on standard benchmarks. All metrics reported as of March 2026.

Language	Tokenization Framework
English	Standard Tokenization
Chinese	Standard Tokenization
Dutch	Standard Tokenization
French	Standard Tokenization
Nepali	Bespoke Devanagari Tokenization
Hindi	Bespoke Devanagari Tokenization

Table 1: Supported languages and their respective tokenization frameworks in Celer 2.6.

Language	Llama 3 Tokenizer	Qwen 2.5 Tokenizer	Celer 2.6 Native
English	120	115	118
Nepali (Devanagari)	845	312	296
Hindi (Devanagari)	810	295	288

Table 2: Tokens required to encode a standard 100-word news passage in English and Devanagari scripts.

Model	MMMU	DocVQA	MathVista	AI2D	ChartQA
Celer 2.6 High (27B)	58.4	91.2	62.1	82.4	83.5
InternVL2 (26B)	51.2	92.9	59.4	84.5	84.9
Qwen2-VL (72B)	64.5	94.0	65.2	86.5	88.1
Llama 3.2 Vision (11B)	50.7	85.1	51.5	78.4	76.2

Table 3: Multimodal benchmark performance of Celer 2.6 High vs frontier vision-language models ( 1 ; 3 ; 15 ) .

Model	MMLU	MMLU-Pro	MATH-500	HumanEval	ACUMEN
Celer 2.6 High (27B)	87.1	78.4	83.2	88.6	71.8
SAGE Actus 2.4 (32B)	89.67	75.60	78.87	86.2	71.8
DeepSeek-R1-Distill (32B)	90.8	84.0	97.3	90.2	68.4
Llama 3.3 70B	86.0	68.9	77.0	88.4	71.3
Qwen 2.5 32B	85.3	73.1	82.4	88.2	67.2
Gemma 3 27B-IT	76.9	67.5	74.2	87.8	63.8

Table 4: Evaluation of Celer 2.6 High against SAGE Actus 2.4, frontier reasoning specialists, and Tier L baselines.

Model	MMLU	MMLU-Pro	MATH-500	HumanEval	ACUMEN
Celer 2.6 Mid (10B)	80.2	71.8	73.6	83.8	68.4
Phi-4 14B	84.8	70.1	80.4	82.6	65.2
Qwen 2.5 14B	79.7	65.4	72.1	82.3	63.8
Gemma 3 12B-IT	71.9	58.2	66.4	85.4	61.4
Llama 3.1 8B-Instruct	68.4	44.3	51.9	62.1	54.2
Celer 2.6 Low (5B)	74.8	67.2	69.4	79.2	63.6

Table 5: Evaluation of Celer 2.6 Mid and Low against Tier S baselines. 2 2 2 Llama 3.1 8B-Instruct is intentionally repeated in both blocks as a shared anchor baseline, enabling direct comparison between the Mid and Low blocks without introducing cross-block normalization artifacts.

Figure 2: Needle-in-a-Haystack evaluation across SAGE Celer 2.6 variants and Celer 2.5 High, ensembled over diverse document sources. Context length on X axis, needle position depth on Y axis. Green indicates high recall, red indicates degraded retrieval.

Configuration	Celer 2.6 High (27B)	Celer 2.6 Mid (10B)	Celer 2.6 Low (5B)	Celer 2.5 High (14B)
All context lengths	98.1	95.2	91.8	92.6
Maximum context length	94.0	91.0	88.0	90.0

Table 6: Needle-in-a-Haystack Recall Summary (%)

Figure 3: Comparison of average recall achieved by SAGE Celer models as context length grows.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Category	Refusal Rate (%)	False Positives (%)
Malware / Zero-Day Synthesis	99.1%	0.4%
Bio-chemical Weapon Formulation	99.8%	0.1%
Political Disinformation / Deepfakes	96.4%	1.2%
Sectarian Hate Speech (South Asia)	98.7%	0.8%
Average	98.5%	0.6%

Appendix

Table 7: Adversarial Safety Evaluation (Celer 2.6 High). Metrics indicate the percentage of prompts successfully refused when falling within prohibited categories.

Panel of Figure 4 — Figure 4: Overall results for Trust & Safety multimodal policy red teaming across SAGE Celer 2.6 variants and SAGE Actus 2.4.

Language Model Self-Verification

See all 16 papers

SEA-LION-v4.8: A Technical Report
Sep 16, 2026Adila Aulia, Ahmed Dabeer, Ahn Jeongmi +38Multilingual Language Models LLM Fine-Tuning
David Wang Dawei
SAGE: Multi-Agent Self-Evolution for LLM Reasoning
Mar 16, 2026Yulin Peng, Xinxin Zhu, Chenxing Wei +4LLM Reasoning Self-Training for Language Models
College of Computer Science and Software Engineering, Shenzhen University, China · Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ), China · School of Information Technology, Carleton University, Canada
Native Multilingual Chain-of-Thought Reasoning in Low-Resource Southeast Asian Languages
Aug 1, 2026Sean Gip Lim, William Chandra Tjhi, Hai Leong ChieuNumerical Reasoning in Language Models Multilingual Language Models
1Nanyang Technological University · 2AI Singapore · 3DSO National Laboratories

Sep 16, 2026J/K move · Enter open · S save

SEA-LION-v4.8: A Technical Report

Adila Aulia, Ahmed Dabeer, Ahn Jeongmi, Antonyrex Sajeban+37

David Wang Dawei

Multilingual Language Models LLM Fine-Tuning

Open paper AskPDF arXiv

We introduce Nemotron-SEA-LION-v4.8, a family of Southeast Asian Languages In One Network (SEA-LION) models built upon NVIDIA Nemotron 3. The family includes 30B-A3B and 120B-A12B models, with both continued-pretrained base checkpoints and post-trained variants. We adapt the models using Southeast Asian, reasoning, code, and multilingual parallel data, followed by post-training with supervised fine-tuning and online on-policy distillation. On SEA-HELM, the 30B-A3B model improves the overall SEA score from 46.06 to 51.57, while the 120B-A12B model improves from 49.30 to 63.44. Across seven Southeast Asian languages, we observe broad capability gains with the 120B-A12B model showing broader and more consistent improvements across tasks.

Abstract

Language Model Self-Verification

Explore similar work

SEA-LION-v4.8: A Technical Report

SAGE: Multi-Agent Self-Evolution for LLM Reasoning

Native Multilingual Chain-of-Thought Reasoning in Low-Resource Southeast Asian Languages