The right model for every agent.
Planning, building, testing and reviewing need different strengths. An AI profile decides which model each agent runs, so you pay for capability where it matters and save where it doesn’t. Use one of ours, or build your own.
- FreeZero Cost$0.00 - $1.20 / 1M tokens
- BudgetLow cost, everyday work$0.00 - $1.20 / 1M tokens
- SmartestHighest capability$0.00 - $50.00 / 1M tokens
- QuickFastest turnaround$0.00 - $10.00 / 1M tokens
- LudicrousSpeed over ceremony$0.00 - $8.70 / 1M tokens
- EU OnlyEU data residency$4.18 / 1M tokens
- Low carbonLower estimated CO2e$0.24 - $0.60 / 1M tokens
Seven profiles, one choice per task
Every profile is a tested set of models for each agent role. Set a default for your organization and override it for a single task or plan. Prices are live from the model catalog.
Cost against quality
Select a profile to open it
Output cost per million tokens
EU-hosted models only
Low carbon profile
Larger mark = faster
FreeZero CostGLM-5.3 FlashDeepSeek V4.1 FlashQwen3.7 Flash+2SpeedMediumQualityGood$0.00 - $1.20 / 1M tokensOutput · Input $0.03 - $0.30 / 1M tokens
Coroid routes each task to cost-efficient models at no cost to you. Good for trying Coroid and for lightweight work. The models used may change over time.
Best For
- Evaluating Coroid
- Personal projects
- Lightweight coding tasks
- Zero-budget experimentation
- Input
- $0.03 - $0.30 / 1M tokens
- Output
- $0.00 - $1.20 / 1M tokens
- Models
- GLM-5.3 Flash · DeepSeek V4.1 Flash · Qwen3.7 Flash · MiMo-V2.6-Flash · Jev 1.13
- Est. CO2e
- 927.5 g / 1M
- Estimated operational inference electricity using Ember/eGRID location factors. Excludes training, devices, water, and embodied hardware.
BudgetLow cost, everyday workDeepSeek V4.1 FlashGPT-6 LunaQwen3.7 Flash+1SpeedFastQualityGood$0.00 - $1.20 / 1M tokensOutput · Input $0.03 - $0.30 / 1M tokens
Cost-efficient models for routine implementation, documentation, and refactoring, where the spec is clear and the change is contained.
Best For
- Simple feature implementations
- Documentation and comments
- Code formatting and linting
- Refactoring tasks
- Input
- $0.03 - $0.30 / 1M tokens
- Output
- $0.00 - $1.20 / 1M tokens
- Models
- DeepSeek V4.1 Flash · GPT-6 Luna · Qwen3.7 Flash · Jev 1.13
SmartestHighest capabilityGPT-6 AstraGemini 3.7 FlashGPT-6 Luna+7SpeedMediumQualityOutstanding$0.00 - $50.00 / 1M tokensOutput · Input $0.04 - $10.00 / 1M tokens
The strongest available models, for architecture decisions, hard debugging, and security-critical code where a wrong call costs more than the tokens.
Best For
- Complex architecture design
- Debugging intricate issues
- Security-critical code
- Algorithm optimisation
- Input
- $0.04 - $10.00 / 1M tokens
- Output
- $0.00 - $50.00 / 1M tokens
- Models
- GPT-6 Astra · Gemini 3.7 Flash · GPT-6 Luna · Jev 1.13 · GPT-6 Sol · Opus 5.5 · GLM-5.3 Flash · Grok 4.7 · Opus 5.5 · GLM-5.3
- Est. CO2e
- 927.5 g / 1M
- Estimated operational inference electricity using Ember/eGRID location factors. Excludes training, devices, water, and embodied hardware.
QuickFastest turnaroundGPT-6 SolGemini 3.7 FlashGrok 4.7+4SpeedFastQualityGood$0.00 - $10.00 / 1M tokensOutput · Input $0.03 - $2.00 / 1M tokens
Fast, inexpensive models for small fixes, prototypes, and exploration, when a quick answer is worth more than a deep one.
Best For
- Rapid prototyping
- Quick iterations
- Simple fixes
- Exploration tasks
- Input
- $0.03 - $2.00 / 1M tokens
- Output
- $0.00 - $10.00 / 1M tokens
- Models
- GPT-6 Sol · Gemini 3.7 Flash · Grok 4.7 · DeepSeek V4 Flash 0731 · GPT-6 Luna · Qwen3.7 Flash · Jev 1.13
LudicrousSpeed over ceremonyMiMo-V2.6-Pro-UltraSpeedDeepSeek V4.1 FlashQwen3.7 Flash+2SpeedFastQualityExcellent$0.00 - $8.70 / 1M tokensOutput · Input $0.03 - $4.35 / 1M tokens
The fastest models with relaxed quality gates. Use it for drafts and urgent fixes, and expect to review the output more closely.
Best For
- Emergency fixes
- Maximum velocity
- Draft implementations
- Rapid experimentation
- Input
- $0.03 - $4.35 / 1M tokens
- Output
- $0.00 - $8.70 / 1M tokens
- Models
- MiMo-V2.6-Pro-UltraSpeed · DeepSeek V4.1 Flash · Qwen3.7 Flash · GPT-6 Luna · Jev 1.13
EU OnlyEUEU data residencyMistral Large 4SpeedFastQualityExcellent$4.18 / 1M tokensOutput · Input $1.36 / 1M tokens
Runs only on models hosted in the EU, so prompts and code are processed within EU jurisdiction. Built for teams with GDPR and data-residency requirements.
Best For
- GDPR-sensitive projects
- EU data residency requirements
- Privacy-critical applications
- Enterprise compliance
- Input
- $1.36 / 1M tokens
- Output
- $4.18 / 1M tokens
- Models
- Mistral Large 4
Low carbonLower estimated CO2eNemotron 3 Nano 30BMistral Small 4MiMo-V2.6-FlashSpeedFastQualityGood$0.24 - $0.60 / 1M tokensOutput · Input $0.06 - $0.15 / 1M tokens
Prefers smaller models on lower-intensity grids. Estimates are operational inference only — not a carbon-neutral guarantee.
Best For
- Lower-intensity inference
- Efficient coding agents
- Cleaner-grid routing
- Everyday software work
- Input
- $0.06 - $0.15 / 1M tokens
- Output
- $0.24 - $0.60 / 1M tokens
- Models
- Nemotron 3 Nano 30B · Mistral Small 4 · MiMo-V2.6-Flash
Or assign every agent yourself
Professional and Enterprise organizations create their own profiles. Start from any Coroid profile — its agent models, fallbacks, and policies are copied in — then change whatever the work needs. Every active model in the catalogue is available to pick from.
Professional and Enterprise91 models · 8 providers
- Primary model
- The model that role runs by default.
- Fallback model
- Where the work goes if the primary is unavailable. Optional, per role.
- Reasoning effort
- How hard the model works the problem before it answers.
- Whose key pays
- Your own provider key or Coroid credits — each model in the profile says which.
- Delivery
- Architect · Developer · Quality assurance · Reviewer
- Support
- Web tester · Assistant · Attachment inspector · Dependency analyzer · Explorer
- Automation
- Hook executor · Developer sub-agent
A few roles narrow the list: the assistant needs a conversational model, and the attachment inspector needs one that can see images.
Owners and admins create and edit profiles. Members choose from them in the work they raise.
One default, overridden where it matters
- Organization default
- Every piece of work runs on it unless something says otherwise.
- Per task or plan
- Override the profile on a single task or a phased plan before it starts.
- Per automation
- Hooks and schedules can run on a profile of their own — cheap models for routine watching, capable ones when it matters.
- Locked once it runs
- A task keeps the profile it started with, and finished work keeps the profile it ran with, so its cost and evidence stay readable later.
Chosen on merit, not origin
The catalog spans frontier labs and open-weight models from around the world. Profiles draw on both:
- Frontier models: the strongest reasoning, for architecture, security-critical code, and hard debugging.
- Cost-efficient open-weight models: strong coding at a fraction of the price, for routine work at scale. Use the EU Only profile when data residency matters.