ToolSpace-8B
Code · Evaluation · 8B · 35B-A3B · 744B-A40B
ToolSpace-8B is a Qwen3-8B agent model trained with ToolSpace trajectories and on-policy agent reinforcement learning for long-horizon tool use.
ToolSpace constructs a tool-composition graph from real execution traces, capturing data, state, and retry dependencies. It synthesizes cross-environment tasks through state-aware walks with rollback and keeps trajectories that pass execution-based verification.
Paper: ToolSpace: Scaling Tool-Centric Environment Synthesis for Long-Horizon Agents.
https://huggingface.co/papers/2611.05036
Model information
| Item | Details |
|---|---|
| Backbone | Qwen3-8B |
| Architecture | Dense, 8B parameters |
| Context length | 65,536 tokens |
| Weight format | BF16 Safetensors |
| Training | Supervised fine-tuning + agent RL |
Results
Results from Table 1 of the ToolSpace paper. Avg. is the unweighted mean over the seven benchmarks.
| Model | AutomationBench | SkillsBench | JobBench | τ³-Bench Banking | BFCL V4 | Harvey LAB | OfficeQA Pro | Avg. |
|---|---|---|---|---|---|---|---|---|
| Qwen3-8B | 0.0 | 0.1 | 0.1 | 4.1 | 42.8 | 0.0 | 3.0 | 7.2 |
| ToolSpace-8B (SFT) | 32.3 | 3.9 | 13.1 | 10.3 | 65.3 | 0.0 | 12.0 | 19.6 |
| ToolSpace-8B + Agent RL (this checkpoint) | 34.3 | 4.1 | 15.5 | 11.3 | 67.5 | 0.0 | 12.0 | 20.7 |
Usage
Serve with vLLM and the included chat template:
vllm serve dongguanting/ToolSpace-8B \
--served-model-name ToolSpace-8B \
--max-model-len 65536 \
--enable-auto-tool-choice \
--tool-call-parser hermes \
--reasoning-parser deepseek_r1
The server exposes an OpenAI-compatible endpoint at http://localhost:8000/v1.
Evaluation commands and benchmark configurations are available in the ToolSpace evaluation directory.