Fresh Models, More Firepower
Today we're excited to announce the addition of four new models to Kodo's multi-model lineup. Starting right now, every Kodo user — including those on the free tier — can access:
- meta-llama/llama-4-scout — Meta's efficient frontier model built for speed and tool use
- meta-llama/llama-4-maverick — Meta's powerhouse model for complex reasoning and generation
- deepseek/deepseek-r1 — DeepSeek's acclaimed reasoning model with transparent chain-of-thought
- google/gemma-3-27b-it — Google's instruction-tuned open model punching well above its weight class
These models are available immediately across the Kodo CLI, VS Code, Cursor, and all JetBrains extensions.
Why These Four?
We don't add models for the sake of padding a list. Every model in Kodo earns its spot by excelling at something developers actually need. Here's why each of these made the cut:
Meta Llama 4 Scout
Llama 4 Scout is Meta's latest efficient model, purpose-built for fast, tool-heavy workflows. With its mixture-of-experts architecture and a massive 10 million token context window, it can ingest entire codebases without breaking a sweat. It's ideal for:
- Quick code generation and boilerplate
- Large codebase navigation and understanding
- Tool-calling and agentic workflows where speed matters
$ kodo --model llama-4-scout "add input validation to all API endpoints"Meta Llama 4 Maverick
Where Scout optimizes for speed, Maverick goes deep. This is Meta's top-tier reasoning model in the Llama 4 family — a 400B+ parameter mixture-of-experts model that rivals the best proprietary models on coding benchmarks. Use it when you need:
- Complex multi-file refactoring
- Architectural planning and system design
- Nuanced code review with deep reasoning
$ kodo --model llama-4-maverick "redesign the caching layer for horizontal scaling"DeepSeek R1
DeepSeek R1 has taken the AI world by storm with its transparent reasoning approach. Unlike traditional models that give you an answer, R1 shows you its step-by-step thinking process — making it invaluable for:
- Debugging complex issues where you need to follow the logic
- Algorithmic problem-solving with visible chain-of-thought
- Learning and understanding unfamiliar code patterns
- Code review where reasoning matters as much as the result
$ kodo --model deepseek-r1 "find the race condition in our websocket handler"Watching R1 reason through your code step-by-step is like pair programming with someone who thinks out loud. It doesn't just find the bug — it shows you *why* it's a bug.
Google Gemma 3 27B IT
Gemma 3 27B IT is Google's instruction-tuned open model, and it's remarkably capable for its size. At 27 billion parameters, it's lightweight enough to run efficiently while delivering surprisingly strong results on:
- Day-to-day coding tasks and quick iterations
- Documentation generation and code explanation
- Test writing and boilerplate generation
- Cost-sensitive workflows where you want quality without burning through credits
$ kodo --model gemma-3-27b-it "write comprehensive JSDoc comments for this module"Mix and Match with Model Overrides
The real power of Kodo's multi-model approach is using the right model for the right job. With these additions, you have even more flexibility. Update your .kodo.config to auto-route tasks:
{
"defaultModel": "claude",
"modelOverrides": {
"tests/**": "gemma-3-27b-it",
"algorithms/**": "deepseek-r1",
"infrastructure/**": "llama-4-maverick",
"scripts/**": "llama-4-scout"
}
}Your tests get written by an efficient model that keeps costs low. Your algorithms get the deep reasoning treatment from DeepSeek R1. Your infrastructure code gets Maverick's architectural chops. And your utility scripts get Scout's speed. All automatic, zero friction.
Open Models Matter
You might notice a theme with today's additions: all four are open models. This is intentional. We believe the open model ecosystem is reaching an inflection point where open-weight models genuinely compete with — and in some cases surpass — proprietary alternatives.
This matters for several reasons:
- Transparency — You can understand what the model is doing and how it was trained
- Community — Open models benefit from a global community of researchers and developers
- Independence — No single company controls the future of your development tools
- Cost — Open models drive down costs across the entire ecosystem
Kodo will always support both proprietary and open models. We want you to choose based on capability, not vendor lock-in.
Getting Started
If you're already using Kodo, these models are available right now — no update needed. Just pass the model flag:
$ kodo --model llama-4-scout "explain this function"
$ kodo --model llama-4-maverick "refactor for performance"
$ kodo --model deepseek-r1 "debug this failing test"
$ kodo --model gemma-3-27b-it "add error handling"New to Kodo? Get started in under 30 seconds:
$ npm install -g @eldlabs/kodo-cli
$ kodo init
$ kodo --model deepseek-r1 "walk me through this codebase"What's Next
Our model catalog is growing fast. We're evaluating new models every week and will continue adding the ones that make a real difference to your development workflow. Got a model you want to see in Kodo? Let us know on Discord or Twitter.
Four new models, zero new complexity. That's the Kodo way.