SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI developer platformsAwesome-LLM-Strawberry

Awesome-LLM-Strawberry

Awesome-LLM-Strawberry is a collection of research materials centered on OpenAI o1 (codename Strawberry) and LLM reasoning technology, hosted on GitHub. This repository gathers a large number of papers, blogs, and projects related to reasoning models, and states that it will be continuously updated to track the frontier of LLM reasoning. The content is organized by topic, covering official introductions and technical reports for models such as OpenAI o1/o3, Gemini 2.0 Flash Thinking, DeepSeek R1, the Qwen series, and Kimi reasoning models, as well as papers and engineering practices in directions such as RL (reinforcement learning) training, inference scaling, process reward models, and test-time compute. It is not itself runnable software or a framework, but rather an aggregated resource index, suitable for researchers, engineers, and readers interested in reasoning models to look up original materials by following the clues.

Open website

About this tool

Awesome-LLM-Strawberry (OpenAI o1) is a repository collecting research materials oriented toward the direction of large language model reasoning, focusing on OpenAI o1 (formerly codenamed Strawberry) and the broader field of LLM Reasoning technology. The repository is positioned as a continuously updated resource index for tracking frontier progress in the field of LLM reasoning.

Purpose This repository is used to centrally organize paper, blog, and project links related to reasoning models, making it convenient for readers to quickly locate original materials for a given model or technical direction, such as the OpenAI o1 series, Gemini 2.0 Flash Thinking, DeepSeek R1, the Qwen series, and Kimi reasoning models.

Main collected content According to the repository description and the listed entries, the content can be roughly divided into the following categories:

  • Official releases and explanations: including OpenAI's documentation on reasoning, introductions to the o3 preview and o3 mini, ChatGPT Pro, deep research, etc.; Google DeepMind's Gemini 2.0 Flash Thinking; and Anthropic's article on building effective agents.
  • Third-party analysis articles: such as analyses and interpretations of o1's architecture, training infrastructure, and inference scaling laws, as well as discussions on Strawberry, LM self-talk, reverse engineering OpenAI's o1, and more.
  • Opinions and interview notes: such as Ilya Sutskever's view that AI with reasoning capabilities is harder to predict, notes from the OpenAI o1 team AMA, Noam Brown's sharing on planning capabilities, and Hyung Won Chung's "Don't teach. Incentivize.", etc.
  • Open-source reproductions and models: including Qwen3, QwQ, QvQ, DeepSeek R1, DeepSeek V3, DeepSeekMath-V2, the NVIDIA Nemotron series, Skywork R1V2 / Skywork Open Reasoner 1, as well as O1 Replication Journey, OpenO1, Marco-o1, HuatuoGPT-o1,Projects such as DRT-o1.
  • Reinforcement learning and training frameworks: involves OpenRLHF, REINFORCE++ and REINFORCE++-baseline, SkyRL, EasyR1, ROLL, Dr. GRPO, TinyZero, Logic-RL, Open-Reasoner-Zero, VAPO, etc.
  • Technical reports and papers: such as Qwen3 Technical Report, DeepSeek-R1, DeepSeek-Prover-V2, MiniMax-M1, Kimi k1.5 / Kimi k2, KIMI-VL, Llama-Nemotron, Nemotron-Cascade, Magistral, ERNIE-4.5, LongCat Flash, GLM-4.5, Seed Thinking v1.5, MiMo-V2 Flash, Nemotron 3 Nano, etc.
  • Research on reasoning and RL methods: includes Self-generated data improving reasoning, Process Reward Models That Think, Group Sequence Policy Optimization, ProRL / ProRL V2, BroRL, Sample More to Think Less, Concise Reasoning via Reinforcement Learning, Inference-Time Scaling for Generalist Reward Modeling, JudgeLRM, etc.
  • Multimodal and visual reasoning: such as LMM-R1, R1-V, MM-EUREKA, Thinking with Generated Images, Pixel Reasoner, DeepEyes, Bring Reason to Vision, etc.
  • Other related topics: covers o3's high-score performance on ARC-AGI, o1-mini's cost efficiency, model hallucination (Why Language Models Hallucinate), and discussions on reproducibility of reasoning evaluation, etc.

Usage This repository is a collection of materials; the repository description does not provide code, installation commands, or usage steps. Users typicallybrowse the repository directory and item links to jump to the corresponding papers, blogs, or project pages for reading. The repository states that it will be continuously updated to track the frontiers of LLM Reasoning.

Applicable scenarios Suitable for researchers, engineers, and readers concerned with advances in reasoning models, for systematically finding original materials on o1 and related reasoning techniques; it can also be used to quickly understand representative work in a certain type of method (such as RL training, process rewards, test-time compute, long-context reasoning).

Limitations The repository itself does not provide models, inference services, or executable frameworks, nor does it include installation commands or deployment guides; its content is an index pointing to external materials, and specific functions, fees, and usage conditions must be based on each original material. Some links in the repository are internal or third-party pages, and their accessibility and update status depend on the corresponding sources. Regarding the repository's copyright, license, and maintenance cycle, the materials do not currently provide clear explanations.

Related tools

Browse category