sii-research/InnoSpark3.0-35B-260709

TEXT GENERATIONConcurrent Unit Cost:3Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

InnoSpark3.0-35B-260709 is a 35.1 billion parameter education-enhanced model from sii-research, built upon the NEX 35B model family. It is specifically fine-tuned using a multi-stage RL pipeline with general and education-domain data to excel in educational QA, teaching assistance, and learning companionship scenarios. The model demonstrates strong performance in pedagogical tasks, instruction following, and reasoning, making it suitable for various educational AI applications.

Loading preview...

InnoSpark3.0-35B-260709: An Education-Enhanced LLM

InnoSpark3.0-35B-260709 is a 35.1 billion parameter model developed by sii-research, derived from the NEX 35B model family. It undergoes a specialized training regimen involving Supervised Fine-Tuning (SFT) with a blend of general and education-specific data, followed by a multi-stage Reinforcement Learning (RL) pipeline. This process significantly strengthens its capabilities in reasoning, educational question answering, agentic tasks, and robust instruction following.

Key Capabilities

  • Education-Focused Performance: Achieves a total score of 96.6 on EduBench and 95.76 on Pedagogy Benchmark Multilingual, outperforming Qwen3.6-35B-A3B and Nex-N2-mini in educational metrics.
  • Instruction Following: Demonstrates strong instruction adherence with 93.72 on IF-Eval and 68 on IF-bench.
  • Reasoning: Shows competitive reasoning abilities, scoring 86.36 on GPQA Diamond and 90 on AIME26.
  • Tutoring & Explanation: Optimized for generating explanations, providing scaffolded instruction, and guiding reasoning processes.

Ideal Use Cases

This model is particularly well-suited for:

  • Educational QA and Tutoring: Answering pedagogical questions, providing step-by-step explanations, and supporting homework.
  • Teaching Assistance: Generating lesson plans, creating teaching materials, and assisting with classroom scenarios.
  • Learning Companionship: Developing student-facing dialogue agents and personalized learning support systems.
  • General Assistant Tasks: Capable of general instruction following and agent-style tasks beyond the educational domain.

Finetunes
1 models