Strongest fit
Who it is built for
- Researchers evaluating coding-model game-building capability
- Developers inspecting playable coding-agent outputs
- Teams comparing model progress on a shared game-development benchmark
MirageML Bench
A playable benchmark for coding models building GTA in San Francisco
MirageML Bench tracks how far coding models can progress toward building a GTA-style game set in San Francisco from clean-room repositories. It presents playable submissions, their model and build information, source repositories, deployment links, and stated provenance for direct inspection.
MBMirageML BenchProduct screenshot pending
- Best for
- Researchers evaluating coding-model game-building capability
- Pricing signal
- Not publicly listed
- Primary category
- Developer tools
- Last verified
- Jul 30, 2026
The decision
Should MirageML Bench make your shortlist?
Start with the job, the team, and the constraints. Product fit becomes much clearer when those three line up.
Practical use cases
Jobs it can take on
- Play an accepted coding-model game submission
- Inspect a submission's repository and live deployment
- Submit a benchmark entry through the project's GitHub issue template
Before you choose
Know the tradeoffs
- The benchmark is focused on a specific GTA-style game-building challenge rather than general software development.
- The currently displayed Bayline Heat submission is marked as having unverified lineage because its prior build process did not preserve clean-room provenance.
Inside the product
What you can actually do with it.
The core product capabilities, grouped around the work they enable.
Playable submissions
The benchmark presents accepted submissions with a direct Play link so visitors can inspect the resulting game.
Model and build details
Each listed submission identifies the coding model and the date on which the build was completed.
Source and deployment links
Submission details include links to the source repository and the deployed game when available.
Typical workflow
Coding-model game builds are submitted to the Mirage project for inclusion in the benchmark. Accepted entries are shown in a playable deck with implementation, source, deployment, and lineage details. Visitors can play the submission and inspect the linked project materials.
Plans and official links
Plans and accessNot publicly listed.
See the entry price, free access options, company details, and direct vendor destinations in one place.
Starting pointNot publicly listed
Official destinations
Choosing for a real workflow?
Make the tool work with the rest of your operation.
We map the workflow, connect existing systems, choose what to buy, and build what is missing.