Introduction
Vibe CLI ships with three default models, but they are not the best for coding. If you are using Vibe for actual software engineering, you want smarter, faster, and cheaper models configured properly.
This guide covers:
- Which Mistral models to use for coding and agentic tasks
- How to add them to your
~/.vibe/config.toml - Essential skills that make Vibe actually useful for developers
Recommended Models for Coding
| Model | Alias | Input $ | Output $ | Best For |
|---|---|---|---|---|
| mistral-small-2603 | small4 | $0.2/M | $0.6/M | Daily coding, fast autocomplete, simple refactors |
| mistral-medium-2508 | medium | $1.5/M | $4.5/M | Complex logic, debugging, architecture decisions |
| mistral-large-2512 | large | $2.0/M | $6.0/M | Deep reasoning, legacy code analysis, hard bugs |
| devstral-small-latest | devstral-small | $0.1/M | $0.3/M | Cost-conscious coding, high-volume agent tasks |
| devstral (local) | local | Free | Free | Privacy-first, offline work, zero API costs |
My recommendation: Use small4 as your daily driver. It is the sweet spot for coding—fast, cheap, and good enough for 90% of tasks. Switch to large only when you are stuck on a hard problem.
How to Add Models
Vibe uses a TOML config file. The devstral models are already configured by default. You just need to add the large, medium, and small models next to them.
Step 1: Open the file:
~/.vibe/config.toml
Step 2: You will already see the devstral models in there:
[[models]]
name = "devstral-small-latest"
provider = "mistral"
alias = "devstral-small"
temperature = 0.2
input_price = 0.1
output_price = 0.3
thinking = "off"
auto_compact_threshold = 200000
[[models]]
name = "devstral"
provider = "llamacpp"
alias = "local"
temperature = 0.2
input_price = 0.0
output_price = 0.0
thinking = "off"
auto_compact_threshold = 200000
Step 3: Paste these models right below the devstral ones:
[[models]]
name = "mistral-small-2603"
provider = "mistral"
alias = "small4"
temperature = 0.2
input_price = 0.2
output_price = 0.6
thinking = "off"
auto_compact_threshold = 200000
[[models]]
name = "mistral-medium-2508"
provider = "mistral"
alias = "medium"
temperature = 0.2
input_price = 1.5
output_price = 4.5
thinking = "off"
auto_compact_threshold = 200000
[[models]]
name = "mistral-large-2512"
provider = "mistral"
alias = "large"
temperature = 0.2
input_price = 2.0
output_price = 6.0
thinking = "off"
auto_compact_threshold = 200000
Step 4: Save the file and restart Vibe.
Now switch models inside the CLI:
vibe > /model small4
Switched to small4
Essential Skills for Coding
Skills turn Vibe from a chatbot into a specialized coding agent. Here are the ones worth using:
| Skill | What It Does | When to Activate |
|---|---|---|
| architecture-diagram | Generates SVG architecture diagrams | Designing systems, documenting infra |
| docker-management | Manages containers, images, Compose stacks | Docker workflows, deployments |
| frontend-design | Builds accessible web interfaces | UI components, CSS reviews |
| github-code-review | Reviews code with inline comments | Pre-commit checks, PR feedback |
| systematic-debugging | 4-phase root cause investigation | Complex bugs, production issues |
| test-driven-development | Enforces Red-Green-Refactor TDD | New features, bug fixes |
| writing-plans | Creates implementation plans | Sprint planning, feature breakdowns |
| requesting-code-review | Static scans before committing | Final check before prod push |
| oss-forensics | Investigates supply chain attacks | Security audits, dependency checks |
| popular-web-designs | 54 production-ready design systems | Prototyping, UI inspiration |
How to Install Skills
Skills live in ~/.vibe/skills/:
mkdir -p ~/.vibe/skills
Clone the skills repository:
git clone https://github.com/DevAgarwal2/vibe-skills.git ~/.vibe/skills
Enable All Skills in Config
Open your Vibe config file:
~/.vibe/config.toml
Find the enabled_skills line and change it to include all skills:
# BEFORE:
enabled_skills = ["frontend-design"]
# AFTER:
enabled_skills = ["frontend-design", "test-driven-development", "systematic-debugging", "writing-plans", "requesting-code-review", "popular-web-designs", "docker-management", "github-code-review", "architecture-diagram", "oss-forensics"]
Save the file. All skills are now active by default.
Activate a Skill
Inside Vibe:
vibe > /skill test-driven-development
Activated: test-driven-development
Use Multiple Skills
vibe > /skill writing-plans,test-driven-development
Activated: writing-plans, test-driven-development
Daily Workflow
Here is how I use Vibe for actual work:
1. Start with the cheap model:
vibe > /model small4
2. Plan the feature:
vibe > /skill writing-plans
vibe > Plan a Stripe payment integration for our API
3. Implement with TDD:
vibe > /skill test-driven-development
vibe > Implement task 1: create the Payment model
4. Review before push:
vibe > /skill requesting-code-review
vibe > Review all changes in this session
5. Stuck on a hard bug? Switch up:
vibe > /model large
vibe > /skill systematic-debugging
vibe > This race condition only appears under load...
Cost Cheat Sheet
| Task | Model | Estimated Cost |
|---|---|---|
| Autocomplete / small refactor | small4 | ~$0.01 |
| Feature planning | small4 | ~$0.05 |
| TDD implementation | small4 | ~$0.10 |
| Code review | devstral-small | ~$0.02 |
| Complex debugging | large | ~$0.50 |
| Architecture design | large | ~$0.30 |
| Daily 4-hour session | small4 | ~$2–4 |
Quick Commands
| Command | Action |
|---|---|
/model small4 | Switch to fast/cheap model |
/model large | Switch to smart model |
/model local | Use local devstral (free) |
/skill <name> | Activate skill |
/skills | List installed skills |
/clear | Reset conversation |
Bottom Line
- small4 = daily driver for 90% of coding
- large = backup for hard problems
- devstral-small = when you want to save money
- local = when you need privacy or offline work
- Skills = what makes Vibe worth using over plain chat
Configure once, code forever.