ByteDance open-sources Axion 1.0-35B agent foundation model with full reproducible recipe
pmttyji · reddit · 2026-11-04
Axion 1.0 is a fully open 35B agent foundation model trained with a remarkably simple recipe: two lightweight SFT stages plus RL — no mid-training, DPO, or on-policy distillation — yet it's competitive with leading commercial models on several agentic benchmarks.
- Stage one builds self-contained reasoning via model-aware data construction: synthesizing hard problems and adversarially strengthening them from the base model's own successful strategies.
- Stage two extends to tool-dependent reasoning with a scalable pipeline of diverse, automatically verifiable tasks grounded in real-world data; RL then strengthens long-horizon reasoning and tool use.
- Open beyond weights: data construction, generation, filtering, verification, prompts, and training code all released (GitHub under bytedance, arXiv:2611.03233), including the training dataset.
Related event: ByteDance and Partners Open-Source Axion 1.0, a 35B Agent Model(2 posts)→
More from coding & agent
- 网友用多 Agent 循环工作流筛出 1132 个 B 站神级视频建站 — huangyun_122 · 2026-11-04
- Graphic designer builds NASA-data 3D Solar System in 2 months with Claude Code, no coding — Shevei1342 · 2026-11-04
- How to Run a 5-Person Office With Claude: Assistant, Bookkeeper, PM, Analyst & Office Manager — evielync · 2026-11-04
- Replit's User Ops lead built an AI app to prioritize and route support tickets — amasad · 2026-11-04
- Autonomous agent designs target-specific MSAs to improve AlphaFold 3 complex prediction — SJTU · 2026-11-04
- NVIDIA paper: frequent context compaction preserves accuracy, long-context models may be unnecessary — nvidia · 2026-11-04