How Does It Work?
Weco autonomously generates and tests candidate solutions, keeping improvements and discarding failed attempts. Click any node to inspect the code.
Example Use Cases
Weco works best when you have well-scoped code to optimize against an efficient eval pipeline. Here are some examples:
Prompt & VLM Optimization
Improve accuracy in vision-language pipelines. Weco iteratively improves prompts, model calls, and post-processing logic until accuracy improves across a held-out set.
See more prompt optimizationsCUDA Kernel Optimization
Reduce GPU inference latency without changing outputs. Weco proposes and tests kernel-level changes until performance improves under the same correctness checks.
See more performance optimizationsKaggle & Benchmark Optimization
Push leaderboard scores with systematic iteration. Weco tests modeling, feature, and code changes until performance improves consistently across runs.
See more benchmark optimizationsAlgorithmic Pricing & Bidding
Optimize sequential pricing and bidding strategies in competitive markets. Weco evolves decision logic until profit and win-rate improve under realistic competition.
See pricing & bidding examplesRouting & Planning Systems
Solve complex vehicle routing, delivery scheduling, and logistics planning problems. Weco iterates on heuristic algorithms until routes are faster, cheaper, and more reliable.
See routing & planning examplesScientific ML Optimization
Optimize ML models for molecular behavior, stability, or efficacy. Weco tests changes until predictions improve against real experimental data.
See more scientific optimizationsBuilt by Frontier AI Researchers
We're the team behind AIDE, which achieved ~4× the medal rate of the next best autonomous agent across 75 Kaggle competitions on OpenAI's MLE‑Bench. Independently validated by researchers at OpenAI, Meta, and Sakana AI.
Benchmark Performance(Medal Rate on MLE-Bench)
Academia and Industry Recognition
Improvements You Can Actually Ship
Ship breakthroughs overnight
Weco can run for weeks without any human intervention.
Optimized for cost efficiency
Each candidate costs fractions of a cent. Find the non-obvious wins that manual iteration misses.
Your data never leaves your machine
Eval code runs locally where your data lives. Only metrics and diffs are sent - review the open-source CLI to verify.
Works with any language
Python, C++, Rust, JS - if it prints a metric to stdout, Weco optimizes it.
See every experiment in one tree
Each run produces a searchable tree of candidates. Compare any two nodes side-by-side.
Steer with natural language
Add constraints like "avoid unsafe memory access" or "prioritize readability" to guide the search.
Trusted by frontier AI labs
What are you trying to do?
I want to build my own AI Scientist
AIDE ML
- Reference implementation of the AIDE algorithm.
- Ideal for academic and industrial researchers to build on and publish.
I want to improve my code
Weco Platform
20 credits free (≈ 100 steps)
- Works with your local evaluation pipeline.
- Steer experiments with natural language.
- Web dashboard for real-time observability.
Frequently Asked Questions
“So amazing to see something built by this team that's substantially underpinning and influencing OpenAI's agentic roadmap.”
Start Optimizing in Minutes
Point Weco at your eval script, run an optimization, and ship winning code: