GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
Spreadsheet-RL
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
Spreadsheet-RL
Model Tree
Qwen/Qwen3-8B
Input Modalities
Output Modalities
Supported Functionality
Spreadsheet-RL/Spreadsheet-RL-8B.| Field | Value |
|---|---|
| Base model | Qwen/Qwen3-8B |
| Released checkpoint | Step 35, the best observed checkpoint |
| Training method | GRPO with outcome-based rewards |
| Environment | Spreadsheet Gym with Microsoft Excel 365, spreadsheet-native tools, SandboxFusion code execution, and async Excel recalculation/reward service |
| Training data | Spreadsheet-RL training split: 5,928 filtered ExcelForum tasks |
| Evaluation | SpreadsheetBench |
| License | Apache-2.0, following the base model license |
For full details, please see the paper. The released 8B checkpoint uses:
| Hyperparameter | Value |
|---|---|
| Algorithm | GRPO; KL-regularized against a frozen reference model; 1 PPO epoch |
| Released checkpoint | 35 RL steps |
| Prompt/response limits | 8,192 / 24,576 tokens |
| Rollout sampling | temperature 0.6; top-p 0.95; top-k 20 |
| Batching | 128 prompts/step; 8 rollouts/prompt; 1,024 rollouts/step |
| Training sampler | Difficulty-aware mixture with difficulty level 5 excluded |
| Optimizer | AdamW; learning rate 2e-6; weight decay 0.01; constant schedule |
| KL loss | low-var KL; coefficient 0.001 |
Spreadsheet-RL improves Qwen3-8B through spreadsheet-native interaction design, comprehensive tool access, and RL post-training.
| Benchmark | Base | + Agent Harness & Full Tools | Spreadsheet-RL-8B (step 35) |
|---|---|---|---|
| SpreadsheetBench Pass@1 | 15.9 | 16.7 | 22.3 |
The released step-35 checkpoint is the best observed checkpoint and improves the full-harness pre-RL result by 5.6 percentage points.
The corresponding 4B release starts from Qwen/Qwen3-4B-Thinking-2507. Its 2507 checkpoint postdates the original Qwen3 series, including Qwen/Qwen3-8B, so model size and base-checkpoint vintage are confounded. Under the full Spreadsheet-RL harness, the 8B model trails the 4B model both before RL (16.7 versus 19.3) and after RL (22.3 versus 23.4); this should not be interpreted as a clean size-only comparison. Instead, the 8B gain from 16.7 to 22.3 shows that Spreadsheet-RL remains effective at a larger model scale, and we release the post-RL checkpoint to make that scaling result reproducible.
Install the standard Transformers stack and load the checkpoint:
python
from transformers import AutoModelForCausalLM, AutoTokenizermodel_id = "Spreadsheet-RL/Spreadsheet-RL-8B"tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)model = AutoModelForCausalLM.from_pretrained(model_id,torch_dtype="auto",device_map="auto",trust_remote_code=True,)
For task evaluation and agent rollouts, use the full Spreadsheet-RL codebase with the released dataset and Spreadsheet Gym:
bash
hf download Spreadsheet-RL/Spreadsheet-RL --repo-type dataset --local-dir datagit clone https://github.com/Spreadsheet-RL/Spreadsheet-RL.git
The default training/evaluation harness is maintained in the code repository under configs/, scripts/, reward/, and verl/.
bibtex
@misc{chi2026spreadsheetrl,title = {Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning},author = {Banghao Chi and Yining Xie and Mingyuan Wu and Jingcheng Yang and Jize Jiang and Zhaoheng Li and Shengyi Qian and Minjia Zhang and Klara Nahrstedt and Rui Hou and Xiangjun Fan and Hanchao Yu},year = {2026},eprint = {2605.22642},archivePrefix = {arXiv},primaryClass = {cs.AI},doi = {10.48550/arXiv.2605.22642},url = {https://arxiv.org/abs/2605.22642}}
| Actor update batching | mini-batch 64; dynamic batch sizing enabled |
| Hardware | 2 nodes x 4 GPUs |