AI models
SeoFreshUp lets you pick a different AI model for each role so you can balance cost vs. quality per task. Cheap-and-fast for bulk audits; high-quality for cornerstone rewrites.
Roles
| Role | Default | Used for |
|---|---|---|
| Standard | openai/gpt-4o-mini | AI verdict (audit), duplicate detection, link suggestions |
| Deep | (empty = uses Standard) | Override if you want higher quality risk analysis |
| Rewrite | anthropic/claude-haiku-4.5 | Article rewrite (long-form) |
| Vision | (empty = uses Standard) | Image quality analysis (when implemented) |
| Image generation | google/gemini-3-pro-image-preview | Featured image + info-card image generation |
Curated list
The dropdown shows a curated selection of models we’ve tested and confirmed compatible. As of May 2026:
Standard / Deep (analysis)
| Model | Cost (per 1M input) | Speed | Notes |
|---|---|---|---|
openai/gpt-4o-mini | $0.15 | Fast | Default, very cheap, decent quality |
openai/gpt-5-mini | $0.20 | Fast | Slightly better than gpt-4o-mini |
anthropic/claude-haiku-4.5 | $1.00 | Fast | Better quality, ~7× cost |
google/gemini-2.5-flash | $0.075 | Very fast | Cheapest, OK for bulk |
:online variants (e.g. perplexity/llama-3.1-sonar-online) | $1.00+ | Slow | Real-time web search, useful for risk analysis |
Rewrite (long-form)
| Model | Cost (per 1M output) | Speed | Notes |
|---|---|---|---|
anthropic/claude-haiku-4.5 | $5 | Fast (~30s for 1500-word article) | Default — good balance |
anthropic/claude-sonnet-4.6 | $15 | Medium (~60s) | Best quality, recommended for cornerstone articles |
anthropic/claude-opus-4.7 | $75 | Slow (~120s) | Overkill for most blogs |
openai/gpt-4o-mini | $0.60 | Very fast (~10s) | Cheap but less polished |
openai/gpt-5 | $30 | Medium | Quality on par with Claude Sonnet |
Image generation
| Model | Cost per image | Notes |
|---|---|---|
google/gemini-3-pro-image-preview | ~$0.04 | Default — high quality, native 16:9 |
google/gemini-2.5-flash-image (Nano Banana) | ~$0.02 | Faster, slightly lower quality |
| Custom OpenRouter image model | varies | Type the model ID in the “Custom” field |
Custom model
If you want to use a model not in the curated list:
- Switch the dropdown to Custom
- Paste the OpenRouter model ID (e.g.,
mistralai/mistral-large-2) - Save
The plugin will use it without validation. If the model ID is invalid, you’ll get an error from OpenRouter at the next AI call.
Discovering live models
OpenRouter adds/deprecates models monthly. To see what’s currently available, click 🔍 Discover available models in Settings → API Keys & Models → Image generation. This queries the live OpenRouter catalog and lists models with image output capability.
For text models, browse openrouter.ai/models.
When to override the default
- You write technical articles (dev, science): switch Rewrite to
claude-sonnet-4.6for better technical accuracy - You write quick news/opinion pieces: keep
claude-haiku-4.5(default), it’s enough - You operate a budget blog: switch Standard to
google/gemini-2.5-flash(cheapest) - You need real-time web data for risk analysis: use a
:onlinevariant for Deep model
Switching models without re-auditing
When you change a model, existing analyzed articles don’t get re-analyzed automatically. To re-analyze with the new model:
- Per article: ⚡ Action → 🤖 Re-analyze
- Bulk: select rows → Bulk action → Re-analyze
- Full blog: 📊 Analyse complète (will re-run all AI calls — costs $$)