AI Model Strategy for OpenCode
A small team of models beats constant switching. This page lays out exactly which model handles which job — for building apps and for running a portfolio of 100 Astro blogs and lead-generation sites — plus the workflows, rules, and workload math behind the assignments.
Two Game Plans, One Principle
The two jobs — building apps and running a 100-site portfolio — need different lineups, but the same discipline: the right model for the risk level of the task.
🏗️ Building Apps in OpenCode
- Claude Opus 5.5 leads architecture and audits — high cost of being wrong.
- GPT-6 is the primary developer — implements features, tests, debugs.
- MiMo 2.6 handles cheap, clearly scoped work — Pro for components, Flash for formatting.
🌐 Running 100 Astro Sites
- Claude Opus 5.5 owns the master theme and major redesigns only.
- GPT-6 owns hard development, research, articles, and audits.
- MiMo 2.6 does the bulk of routine site and content work.
Building Apps: The Team
Six roles, three models. OpenCode supports different models for different agents, so these roles can be configured once instead of switched by hand every time.
| Job | Preferred model | When to use it |
|---|---|---|
| Lead architect | Claude Opus 5.5 | Plan architecture, uncover dependencies, challenge technical assumptions |
| Primary developer | GPT-6 (coding-capable OAuth model) | Implement features, edit files, write tests, debug |
| Second-opinion reviewer | Claude Opus 5.5 | Audit completed features, identify flaws and security issues |
| Budget developer | MiMo 2.6 Pro | Straightforward components, repetitive coding, well-defined changes |
| Cheap utility work | MiMo 2.6 Flash | Formatting, documentation, simple refactors |
| Complex bug investigator | GPT-6 High reasoning, if available | Trace difficult failures and propose targeted fixes |
Building Apps: The Workflow
Plan Claude Opus 5.5
Research, architecture, edge cases, acceptance criteria.
Build GPT-6
Implement one feature at a time, run tests.
Audit Claude Opus 5.5
Review correctness, security, and missing requirements.
Cleanup MiMo 2.6 Pro
Documentation, small fixes, routine components.
Golden Rules
Rule 1 Independent judgment
The model that writes the code should not be the only model deciding whether it is correct.
Rule 2 Evidence over assumptions
No model should substitute its own assumptions for research or actual testing.
Rule 3 Verify cheap work
Use MiMo for work where correctness is easy to verify. Cheap generation is not cheap if it creates hours of rework.
Rule 4 No prestige assignments
Do not pay for Opus-level reasoning to generate every blog post — and never let a budget model redesign the master theme.
Portfolio: The Four Priorities
1 · Consistency
Every site follows the same Astro theme, component rules, and technical standards.
2 · Search performance
Sites are structured to earn search visibility, including traditional search and AI-powered search experiences.
3 · Content quality
Articles are accurate, useful, distinctive, and appropriate for their niche.
4 · Scalability
You can build and maintain 100 sites without spending a fortune on model usage or constantly repairing things.
Theme as Product, Sites as Instances
One major architectural decision prevents more drift than simply choosing a better model: treat the Astro theme as a product and each website as an instance of that product. Models should not be independently designing 100 websites — they should be configuring and populating one well-maintained system.
What each site supplies
- Brand configuration
- Location and business information
- Service data
- Content
What the master theme owns
- Shared, versioned theme components — not 100 independently modified copies
- Astro content collections for structured content and schema validation
- Separate GitHub repositories and Cloudflare deployments per site still work fine
Portfolio: Model Assignments
Roles to start with in OpenCode, then benchmark against your actual sites.
| Responsibility | Primary model | Secondary / budget model |
|---|---|---|
| Master Astro theme architecture | Opus 5.5 | GPT-6 |
| New site setup from template | MiMo 2.6 Pro | GPT-6 |
| Complex Astro / React development | GPT-6 | Opus 5.5 |
| Technical SEO architecture | Opus 5.5 | GPT-6 |
| Keyword and topic research strategy | GPT-6 | MiMo 2.6 Pro |
| Content briefs and outlines | MiMo 2.6 Pro | GPT-6 |
| Long-form SEO articles | GPT-6 | MiMo 2.6 Pro |
| Local service and landing pages | MiMo 2.6 Pro | GPT-6 |
| Content fact-checking and editorial review | GPT-6 with external sources | Opus 5.5 |
| Technical SEO audits and difficult fixes | GPT-6 | Opus 5.5 |
| Routine content updates and metadata | MiMo 2.6 Flash / Pro | GPT-6 |
| Monthly SEO performance analysis | GPT-6 | MiMo 2.6 Pro |
| Major site redesign or template upgrade | Opus 5.5 | GPT-6 |
Search, Answer Engines, and AI Visibility
| Area | What the system should do |
|---|---|
| Technical SEO | Crawlability, indexability, canonical tags, redirects, sitemaps, performance |
| On-page SEO | Search intent, headings, titles, internal links, relevant entities |
| Local SEO | Service-area relevance, legitimate business details, location-specific content |
| Answer Engine Optimization (AEO) | Answer real user questions directly and accurately |
| Generative Engine Optimization (GEO) / AI search | Original, credible, well-structured information that AI systems can cite |
| Structured data | Valid, applicable Schema.org markup matching visible content |
| Content freshness | Update outdated information based on evidence and performance |
| Conversion SEO | Turn qualified visitors into leads, subscribers, or customers |
Automating Ongoing Content
This is where MiMo earns its keep. Automation should create drafts and surface evidence — not blindly publish everything.
Research
Search intent, competitors, sources, content gaps.
Content brief
Outline, audience, questions, internal-link opportunities.
Draft
Write to a defined brief and brand voice.
Editorial review
Check accuracy, originality, unsupported claims, and usefulness.
Publish
Validate frontmatter, links, schema, and build.
Monitor
Analyze Search Console trends and prioritize updates.
Three Reusable Skills
Rather than unique instructions for every website, build three shared OpenCode skills. They share standards but keep separate responsibilities, which makes them easier to improve and test.
🏗️ Astro Site Builder
Creates and updates websites while enforcing the master theme, design system, SEO standards, and deployment rules.
MiMo routine site creation GPT-6 exceptions
📝 SEO & Content Engine
Researches keywords and search intent, creates content briefs, produces articles, checks claims, handles internal linking, and prepares publishing-ready content.
📊 Portfolio SEO Auditor
Reviews Search Console data, detects technical problems, identifies declining pages and content opportunities, and prioritizes work across the entire portfolio.
Workload Split
Where the model usage should end up after testing: most routine work on MiMo, important content and SEO decisions on GPT-6, Opus reserved for architecture and hard audits.
Illustrative workload distribution
Illustrative target, not measured token usage or cost.
Who holds the primary role across the 13 portfolio jobs
Counted from the model assignments table above. Opus holds few jobs but the expensive ones: theme architecture, technical SEO architecture, and major redesigns.
Bottom Line
Guidance in this page references Google for Developers documentation on AI features in Search and scaled content abuse. · Google Search Central documentation