[!IMPORTANT] Phased release. Domyn Edge is a rolling release in three bundles. This teacher is part of the second bundle, together with the other domain teachers and their post-training data. See Open resources for everything released so far.
[!Note] This repository holds the agentic teacher of Domyn Edge: Domyn Edge SFT after an agentic supervised pass on varied agentic tasks. It is one of the domain teachers that score the student's rollouts token by token in the multi-teacher on-policy distillation (MOPD) stage of our post-training protocol. It is released for research and reproducibility; it is not a general-purpose assistant.
Domyn Edge Teacher Agentic is a 4B-parameter dense transformer with the same architecture and tokenizer as Domyn Edge SFT. It was trained via Agentic SFT: one epoch at 128k context on software-engineering, terminal and tool-calling data.
The weights are stored in bfloat16.
The model loads and serves exactly like Domyn Edge SFT; see the Domyn Edge SFT card for the recommended environment, Transformers and vLLM instructions.
Chat template. The repository ships the Domyn Edge SFT chat template unchanged; it is the template used during GRPO training and in the evaluation below. It opens the system turn with a fixed identity line ("You are Domyn Edge, a model trained by Domyn.").
Output format. The model reasons first and calls tools when needed, often over several turns.
On our agentic evaluation suite a task took from a few thousand to about 40k tokens across turns,
so leave room in max_new_tokens.
Scores of this teacher next to those of the student, Domyn Edge SFT, with one benchmark per area, to check what the teacher keeps outside its training domain. All scores are percentages.
| Group | Benchmark | Metric | Domyn Edge SFT | Teacher Agentic | Δ |
|---|---|---|---|---|---|
| Math | AIME 2026 | avg@10 | 62.7 | 62.0 | -0.7 |
| Multilingual math | MGSM (de, en, es, fr, ru) | accuracy | 85.8 | 87.1 | +1.3 |
| Knowledge and reasoning | GPQA Diamond | pass@1 | 56.3 | 58.3 | +2.0 |
| Instruction following | IFBench | prompt strict pass@1 | 24.1 | 23.7 | -0.4 |
| Function calling | BFCL v4, without web search (overall) | pass@1 | 36.5 | 60.0 | +17.2 |
| BFCL v4 (non-live) | pass@1 | 51.9 | 45.9 | -6.0 | |
| BFCL v4 (live) | pass@1 | 75.9 | 75.6 | -0.3 | |
| BFCL v4 (multi-turn) | pass@1 | 35.9 | 42.5 | +6.6 | |
| BFCL v4 (memory) | pass@1 | 27.3 | 25.2 | -2.1 | |
| Agentic | SWE-bench Verified | pass@1 | 33.6 | 40.1 | +6.5 |
| Code | LiveCodeBench (full set)* | pass@1 | 73.0 | 70.5 | -2.5 |
*LiveCodeBench (full set) is all 1,055 problems of LiveCodeBench v6, released between May 2023 and April 2025.
Evaluation settings| Benchmark | Metric | Teacher Agentic | Qwen3.5 4B | Δ |
|---|---|---|---|---|
| τ²-bench | accuracy | 52.4 | 68.2 | -15.8 |
| τ³-bench | accuracy | 4.6 | 7.0 | -2.4 |
| SWE-bench Verified | accuracy | 40.1 | 41.3 | -1.2 |
| Terminal-Bench | accuracy | 4.5 | 15.7 | -11.2 |
| BFCL v4 (multi-turn) | accuracy | 42.5 | 45.1 | -2.6 |
| BFCL v4 (memory) | accuracy | 25.2 | 31.6 | -6.4 |
| BFCL v4 (web search) | accuracy | 40.0 | 63.5 | -23.5 |
| HLE | accuracy | 13.2 | 23.0 | -9.8 |
A teacher, not an assistant. This model is one of the domain teachers of multi-teacher on-policy distillation (MOPD), the post-training stage in which the Domyn Edge student generates its own responses and each teacher scores them token by token in its own domain.
Long generations. On hard problems the model can reason for tens of thousands of tokens, and a small share of responses degenerate when they run far past 64k tokens. Cap the generation length.
Safety work is incomplete. No dedicated safety tuning or red teaming was done on this model. Run generated code in a sandbox.
domyn/Domyn-Edge-Teacher-Agenticdomyn/Domyn-Edge-Teacher-Agentic-Mixdomyn/Domyn-Edge-SFT, and its mix domyn/Domyn-Edge-SFT-MixThis model is released under Apache 2.0.
@misc{domynedge2026,
title = {Domyn Edge},
author = {Domyn},
year = {2026},
note = {Technical report, forthcoming in November 2026.},
}
domyn/Domyn-Edge-SFT