Model choice. A complete building workflow.
Bring the model into a workspace with coding agents, preview, data and publishing.
Start a project- Choose your model
Inspect providers, tools and context before building.
- Choose your coding agent
Codex and OpenCode runtimes, subject to configuration.
- Keep project source
Push workspace changes to GitHub for review and handoff.
- Fund usage as needed
A metered wallet without a recurring builder subscription.
Compare reviewed model options
Live catalog entries matched to exact publisher licenses. Compare the model, then open its full details.
| Model | Weight license | Context | Tool use | Input / output |
|---|---|---|---|---|
| Qwen2.5 Coder 32B Instructqwen/qwen-2.5-coder-32b-instruct | Apache-2.0Publisher reviewed | 32.8K tokens | Not listed |
|
These are reviewed examples, not a complete list of open-weight models. Unreviewed licenses remain unknown. Tool support is catalog metadata; test the selected coding runtime. Provider prices are not final RiverX build quotes.
Three terms. Different decisions.
Choose the model, its license and its hosting deliberately.
| Term | What it means | What to verify |
|---|---|---|
| Open weights | Trained model parameters are available under a release-specific license. | Does not establish a builder's source license or your inference provider's data policy.Example publisher model card |
| Open-source software | Software is distributed under terms permitting use, modification and redistribution. | A separate question from model-weight access. A public RiverX platform license is not established here.Open Source Definition |
| Self-hosted inference | You operate the model-serving infrastructure. | RiverX catalog selection does not establish local inference or a custom model-server connection.RiverX product scope |
Reviewed · Published by RiverX
Choose a model for the task
- Check the exact model ID and publisher license.
- Confirm tool support for repository edits and image input for screenshot-led work.
- Use context size and provider rates as selection inputs, not quality rankings.
Run one complete development loop
- Start with a template or a focused brief and choose an available coding runtime.
- Generate a small change, inspect the diff and run the project's tests.
- Check the preview, then publish and repeat the user journey at the live URL.
Compare the cost of accepted work
- Include failed attempts, retries and review effort in your comparison.
- Use actual wallet usage rather than a provider token price to budget a feature.
- Account for top-up fees, applicable tax and application services separately.
Does open-weight mean unrestricted commercial use?
No. Check the exact release's license, notices and use restrictions. Publisher and family names do not establish the terms for every model.
Does RiverX run open-weight models locally?
This page establishes selection from a hosted catalog, not local inference or support for a self-hosted model endpoint.
Why are only some models labeled with a weight license?
License annotations require reviewed evidence for an exact model version. Unreviewed entries remain unknown rather than inheriting a family-wide label.
Is an open-weight model always the cheapest coding choice?
No. Evaluate actual request cost, successful tool use, retries and the effort needed to accept the generated change.
Put the pieces together
Models, instructions and starting points for your next project.