OPEN MODEL RESEARCH

The open competition
to improve intelligence.

Give a model a better set of weights.
Submit your change. Measure the difference. Move intelligence forward.

Explore the leaderboard ↓Start an experiment

✓ LoRA updates✓ Weight diffs✓ Full checkpoints

MODEL + PATCH = NEXT VERSION

A small change. A measurable difference.

THE LEADERBOARD

Track the frontier.

One submission. Two scores.
Ranked by private test performance.

Private test frontier

Current evaluation

0 verified

The next frontier starts here.

Completed evaluations will appear as points.
The line will track the best private test score.

Current evaluationHow scoring works

Rank	Submission	Submitted by	Validation	Private test ↓	Status
—	Original modelBaseline checkpoint Version ID `0ea62f40-7cd8-4d98-9ebb-484192fa05eb`	Official baseline	—	—	Queued
—	Published weight-space edit reproduction Nov 4, 2026 Version ID `a155dc2f-4075-48a3-9102-e1749f2b352f`	ilblackdragon	—	—	Queued

Scores are out of 100. A dash means no result yet. Only completed private-test evaluations receive a rank.

Understand the scores and evaluation versions +

Validation

The equally weighted average of public benchmark scores. Use these benchmarks to develop and improve your changes.

Private test

One overall score on the selected test-set version. It determines rank. Test questions and individual results stay hidden.

A growing test set

When the test set changes, retained models are evaluated again. Only results from the same configuration are compared; previous results remain in submission history.

YOUR NEXT EXPERIMENT

From weight edit to evidence.

Read the submission spec

01 / START

Choose your parent.

Install the CLI, sign in with GitHub, and pull a registered model version from the leaderboard.

02 / EXPERIMENT

Make your change.

Submit a LoRA, a sparse or dense weight diff, or a full checkpoint. Every change keeps its model lineage.

03 / MEASURE

See the difference.

We verify the weights, then run validation and private tests. Compare your scores with the original model.

Choose your starting point.

Work in your terminal, or hand the experiment to a coding agent.

Choose a model, then a parent checkpoint. Your submission automatically enters that model’s active evaluation.

Python 3.12+ required. The GitHub repository is private during launch preparation; you or your agent will need access.

Read the CLI guide

Selected model: qwen35-08b-v1. Pull the original model or any parent version on its leaderboard.

01 / INSTALL & SELECT

git clone https://github.com/ilblackdragon/deltaleague.git
python3 -m venv .venv
source .venv/bin/activate
pip install ./deltaleague
delta competition list
delta competition init qwen35-08b-v1 --output experiment
cd experiment

Edit the downloaded weights and save them to edited.safetensors. Keep the original model/ directory unchanged. Open a research PR with your code, research trace, and validation findings, then attach its URL with --pr. The CLI does not create the PR. Source-free submissions can omit --pr.

02 / BUILD & SUBMIT

delta patch build --parent ./model \
  --modified ./edited.safetensors --output ./submission
delta validate ./submission
delta login
# Open a research PR first; replace PR_URL with its URL
delta submit ./submission --title "My weight edit" --pr PR_URL
# Use the submission ID returned above
delta submission status <submission-id> --watch

Paste this into your coding agent. You’ll complete GitHub authorization in your browser. Start with a bounded experiment and inspect the results.

RESEARCH BRIEF

Help me run one model-weight improvement experiment on Delta League (https://deltaleague.ai).

1. Read https://github.com/ilblackdragon/deltaleague and its ARTIFACTS.md. Clone the CLI repository, create a Python 3.12+ virtual environment, and install it. If repository access is missing, ask me to grant it. Use `delta login` and wait for me to authorize in my browser; never ask me to paste credentials.

2. Run `delta competition list`. Initialize the selected model with `delta competition init qwen35-08b-v1 --output experiment`, then enter experiment/. This downloads the original checkpoint and public validation data. Preserve model/ and delta-model.json.

3. Read evaluation.json for the pinned evaluation configuration and validation.json for public development examples. Do not try to access hidden tests. Check `delta quota` before scheduling an evaluation.

4. Propose one small, reproducible weight edit and explain the hypothesis. Agree a local compute budget with me before starting; do not rent compute or run an unbounded search. Implement the edit on a copy, save edited.safetensors, and compare against the original using public validation where feasible. Record code, parameters, seeds, and honest findings.

5. Build and validate the patch:
   delta patch build --parent ./model --modified ./edited.safetensors --output ./submission
   delta validate ./submission
   Before submitting, open a research PR in ilblackdragon/deltaleague containing the hypothesis, research trace, reproducible code, parent/evaluation IDs, seeds, and public validation findings. Exclude credentials, private test data, and weight artifacts. The CLI does not create this PR automatically.
   Summarize the change and ask me before submitting it, since evaluation consumes my quota. On approval, run `delta submit ./submission --title "SHORT DESCRIPTION" --pr ACTUAL_PR_URL` once, using the PR you opened. If I explicitly choose a source-free submission, omit --pr.

6. Watch the returned submission ID with `delta submission status ID --watch`, then retrieve `delta submission results ID --json`. Report the submission link, lineage, validation score, and private test score compared with the original on the same evaluation config. Update the research PR with the submission link and verified results when available. Clearly distinguish local estimates, pending results, and verified scores.

The open competitionto improve intelligence.