Choosing the Right Model
Evrykit supports three AI providers and multiple models within each. This article helps you pick the right one for your reporting needs.
Quick Comparison
| Model | Speed | Cost | Best for |
|---|---|---|---|
| Gemini Flash (latest) | Very fast | Free tier / very low | Daily operational reports, quick scans |
| Gemini 2.0 Flash | Fast | Very low | Daily use with latest Gemini architecture |
| Gemini 2.0 Flash Lite | Very fast | Near-zero | High-volume use, simple summaries |
| Gemini 1.5 Flash | Fast | Low | Legacy preference, longer context window |
| Claude Haiku 4.5 | Fast | Low | Conversational analysis, readable summaries |
| Claude Sonnet 4.6 | Moderate | Medium | Deep analysis, management narratives, board packs |
| GPT-4o Mini | Fast | Low | Structured output, cost-effective analysis |
| GPT-4o | Moderate | Medium | Complex multi-step reasoning, financial analysis |
Google Gemini
Gemini Flash (latest)
The recommended default for most teams. The gemini-flash-latest alias always resolves to the most recent Flash model, so you automatically benefit from improvements without changing your settings. The free tier is generous — suitable for a team running AI analysis several times a day across multiple reports.
Gemini 2.0 Flash / 2.0 Flash Lite
Slightly more capable than the previous generation for structured data. Flash Lite sacrifices some depth for near-zero cost — useful if you are running automated batch analysis or have a large team with high report volume.
Gemini 1.5 Flash
A previous-generation model with a very large context window (up to 1M tokens). Useful if you need to feed in extremely large datasets, though most ERP reports rendered on screen are well within the standard limits.
Anthropic Claude
Claude Haiku 4.5
The lightweight Claude model. Faster and cheaper than Sonnet, still notably better at natural language than Gemini Flash for narrative-heavy outputs. A good middle ground when you want readable prose without paying Sonnet prices.
Claude Sonnet 4.6
The best model in Evrykit's lineup for producing high-quality management language — well-structured sections, confident recommendations, and nuanced caveats. Recommended for:
- Monthly management reports that will be read by senior staff
- Board packs where the AI-generated summary will be shared externally
- Exception reports where you need the AI to reason carefully rather than just list facts
- Accounting analysis where precision of language matters
OpenAI
GPT-4o Mini
A cost-optimised model that punches above its weight on structured output tasks. Excellent when you want clean tables, bullet lists, and numbered recommendations without spending on GPT-4o. Recommended for routine operational reports where the output format matters more than depth.
GPT-4o
OpenAI's flagship model. Comparable to Claude Sonnet in quality for most ERP analysis tasks. Some users find GPT-4o particularly strong for financial data and numerical reasoning. Requires a funded OpenAI account.
Recommended Setups by Team Size
Small team (1–5 users)
Configure Gemini Flash (latest) as your default. The free tier will cover most use. Add a Claude Sonnet key for the occasional high-stakes report.
Growing business (5–20 users)
Configure Gemini as default for day-to-day use. Set Claude Sonnet as the go-to for any report that will be shared outside the team. Keep GPT-4o Mini as a fallback if one provider is rate-limited.
Enterprise (20+ users)
Configure all three providers. The default model selector in settings becomes less relevant — train users to select the right tool for the job. Consider upgrading API plans for Gemini and Anthropic to avoid rate limits during peak reporting periods (e.g. month-end close).
A Note on Data Privacy
Your report data is sent to the AI provider's API to generate the analysis. Review the data handling policies of each provider before choosing:
- Google Gemini — ai.google.dev/terms
- Anthropic Claude — anthropic.com/privacy
- OpenAI — openai.com/policies/privacy-policy
If your business handles sensitive personal data or operates in a regulated industry, consult your data protection officer before enabling AI analysis on reports that include personally identifiable information.