Seed-Thinking-v1.5: Advancing Superb Reasoning Models with Reinforcement Learning

2025-04-10

AI Summary

Large language models (LLMs) often exhibit limited reasoning capabilities and poor cross-domain generalization on STEM, programming, and general-purpose tasks. Method: We propose a lightweight sparse Mixture-of-Experts (MoE) architecture (20B activated parameters / 200B total parameters), integrated with reinforcement learning–driven chain-of-thought optimization, multi-stage thought distillation, and alignment training to enable “think-before-answer” reasoning. Contribution/Results: We introduce and open-source two high-quality, human-curated benchmarks—BeyondAIME (focused on advanced mathematical reasoning) and Codeforces (targeting competitive algorithmic programming). Experimental results show state-of-the-art performance: 86.7 on AIME 2024, 55.0 on Codeforces, and 77.3 on GPQA. Moreover, the model achieves an 8% win rate improvement over DeepSeek-R1 on non-reasoning tasks, demonstrating significantly enhanced reasoning accuracy and consistent cross-domain generalization.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We introduce Seed-Thinking-v1.5, capable of reasoning through thinking before responding, resulting in improved performance on a wide range of benchmarks. Seed-Thinking-v1.5 achieves 86.7 on AIME 2024, 55.0 on Codeforces and 77.3 on GPQA, demonstrating excellent reasoning abilities in STEM and coding. Beyond reasoning tasks, the method demonstrates notable generalization across diverse domains. For instance, it surpasses DeepSeek R1 by 8% in win rate on non-reasoning tasks, indicating its broader applicability. Compared to other state-of-the-art reasoning models, Seed-Thinking-v1.5 is a Mixture-of-Experts (MoE) model with a relatively small size, featuring 20B activated and 200B total parameters. As part of our effort to assess generalized reasoning, we develop two internal benchmarks, BeyondAIME and Codeforces, both of which will be publicly released to support future research.

Problem

• Enhancing reasoning models with reinforcement learning for better performance.
• Demonstrating superior reasoning in STEM and coding benchmarks.
• Achieving broader applicability beyond reasoning tasks with MoE architecture.

Innovation

• Reinforcement learning enhances reasoning performance
• Mixture-of-Experts model with scalable parameters
• Generalizes well across STEM and coding tasks