AI and models
Free OpenRouter models, ranked
A tier list of the free OpenRouter models worth using in Bricklayer, ranked by how much real building they can carry.
Each entry says what it’s for, when to reach for it, and when to skip it.
Updated July 2026 · 6 min read
How the tiers work.
Rank tracks how much a model can carry on its own. S handles whole features and hard reasoning, A are strong specialists, B are fast and light. Higher isn’t always better: match the model to the task, and a quick B-tier often beats an S-tier for a one-line change.Tap any model to jump to the details below.
A note on “free.”
OpenRouter’s free catalog changes, and free endpoints can be rate-limited or temporarily unavailable. Treat these as model slugs to try in Bricklayer, then check OpenRouter’s model page for current availability.SDo the heavy lifting
The models you should default to for real building. Strong at multi-step work, tool use, and holding a whole project in context.
Poolside Laguna M.1
poolside/laguna-m.1:freeA coding-agent model built for complex software work, tool use, and long-context tasks. It fits multi-file Luau features and careful repo exploration well, because it can read a lot of the project before it starts editing.
Reach for it when
You are adding or changing a real feature and want the model to explore the code first, then implement it in reviewable steps.
Skip it when
You only need a one-line answer or a tiny script. A smaller B-tier model will turn it around faster.
How to use it: Ask it to inspect the project first, then make one focused feature at a time.
NVIDIA Nemotron 3 Ultra
nvidia/nemotron-3-ultra-550b-a55b:freeA large reasoning and orchestration model to reach for when a sync issue, architecture choice, or stubborn bug needs careful thought rather than a fast guess. It reasons through cause and effect instead of pattern-matching to the nearest fix.
Reach for it when
A bug survived two quick attempts, or you are choosing between two ways to structure a system and want the trade-offs laid out.
Skip it when
The task is straightforward. Its slower, more deliberate answers are wasted on routine edits.
How to use it: Use it for diagnosis and planning; hand the final edit to a faster coding model if you prefer.
AExcellent specialists
Not quite everyday defaults, but the best tool for a specific job. One leans into fast coding, the other into planning and design.
Cohere North Mini Code
cohere/north-mini-code:freeA purpose-built agentic coding model with a small active footprint. Good for fast implementation loops and structured work with developer tools, so it keeps pace when you already know what you want built.
Reach for it when
You have a clear spec and want quick, structured passes: implement, review the diff, adjust, repeat.
Skip it when
The problem is still fuzzy or needs deep reasoning. Hand that to an S-tier model first.
How to use it: Give it context, then ask for a short checklist or a review of a diff.
Google Gemma 4 31B
google/gemma-4-31b-it:freeIts large context window, optional reasoning mode, and native function-calling support make it useful for feature briefs, gameplay loops, and turning a broad idea into small milestones, the thinking that happens before any code gets written.
Reach for it when
You are at the whiteboard stage: pitching a game loop, writing a brief, or breaking a big idea into buildable steps.
Skip it when
You are ready to write and test code. A coding-focused model will implement more reliably.
How to use it: Use it for feature briefs, progression loops, and breaking a big game idea into milestones.
BFast and useful
Small and quick. Not for heavy features, but ideal for fast edits, quick questions, and learning what the code is doing.
OpenAI gpt-oss-20b
openai/gpt-oss-20b:freeAn open-weight mixture-of-experts model optimized for efficient inference. Good for small edits, quick questions, and fast scripting experiments when speed matters more than depth.
Reach for it when
You want a single script tweaked, a short answer, or a fast experiment without waiting on a bigger model.
Skip it when
The change spans several files or needs careful reasoning. Scope creep will trip it up.
How to use it: Keep requests narrow: one script, one behavior, one verification step.
NVIDIA Nemotron Nano 9B V2
nvidia/nemotron-nano-9b-v2:freeA compact model designed to handle both reasoning and direct responses, which makes it handy for walking through logic slowly and explaining code decisions in plain language rather than just handing back an edit.
Reach for it when
You want to understand what a script does or why an approach was chosen before you change it.
Skip it when
You need a large or precise implementation. Its size shows on heavier coding work.
How to use it: Ask it to explain the trade-offs and then request the smallest safe change.
A simple way to pick
Building a feature?
Something confusing?
Shaping the idea?
Browse models on OpenRouter