{"name":"axion1_0","repo":"bytedance/axion1_0","url":"https://midorreal.com/project/bytedance-axion1-0","repository":"https://github.com/bytedance/axion1_0","description":"Frontier-level agent capabilities with nothing more than high-quality data and a simple training recipe: 2 SFT + RL, no mid-training/DPO/OPD. Full recipe released.","facts":{"license":{"value":"Apache-2.0","source":"https://github.com/bytedance/axion1_0","checked":"21 Nov 2026"},"language":{"value":"Python","source":"https://github.com/bytedance/axion1_0","checked":"21 Nov 2026"},"homepage":{"value":"https://axion1-0.github.io","source":"https://axion1-0.github.io","checked":"21 Nov 2026"},"releases":{"value":"0","source":"https://github.com/bytedance/axion1_0/releases","checked":"20 Nov 2026"},"contributors":{"value":"5","source":"https://github.com/bytedance/axion1_0/graphs/contributors","checked":"20 Nov 2026"},"archived":{"value":"no","source":"https://github.com/bytedance/axion1_0","checked":"21 Nov 2026"}},"stars":102.0,"writeup":{"what_it_is":"Axion 1.0 is an open project that releases training code, data construction methods and models for agent foundation models. It describes a training pipeline of two stages of supervised fine-tuning followed by reinforcement learning, built on Qwen base models.","audience":"Researchers and engineers training open agent language models","claims":[{"kind":"specific","claim":"Axion 1.0-122B scores 59.0 on full HLE with restricted website access","excerpt":"Axion 1.0-122B scores **59.0** on full\n  HLE with restricted website access","status":"not_checked"},{"kind":"specific","claim":"Axion 1.0-35B scores 62.2 on SEAL-0","excerpt":"Axion 1.0-35B scores **62.2** on SEAL-0.","status":"not_checked"},{"kind":"specific","claim":"Training uses a three-stage pipeline: two SFT stages followed by RL","excerpt":"Two stages of supervised fine-tuning (SFT) followed\n  by reinforcement learning (RL)","status":"not_checked"},{"kind":"specific","claim":"Stage I uses 10,819 trajectories of scientific and mathematical reasoning","excerpt":"Stage I uses **10,819 trajectories** of scientific and mathematical reasoning.","status":"not_checked"},{"kind":"specific","claim":"The repository includes data generation, two-stage SFT, asynchronous RL, and benchmark grading code","excerpt":"The repository includes data generation and\n  filtering, prompts, two-stage SFT, asynchronous RL, and benchmark preparation\n  and grading.","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-21"},"why_now":null,"owner_supplied":null,"score":{"verdict":"mid","hype":54,"reality":35,"gap":20,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":null,"changes":[{"date":"2026-11-02T04:42:38+00:00","text":"Repository created","source":"https://github.com/bytedance/axion1_0"}],"cite":"axion1_0 is a GitHub project at https://github.com/bytedance/axion1_0 written mainly in Python, described there as \"Frontier-level agent capabilities with nothing more than high-quality data and a simple training recipe: 2 SFT + RL, no mid-training/DPO/OPD. Full recipe released\".","attribution":"Data from Mid or Real, https://midorreal.com/project/bytedance-axion1-0, checked 22 November 2026. Attribution with a link is required.","still_real":null}