1 — LLMs, closed book
Each provider's model answers the exact same question with no retrieval. Differences here are differences in training data and internal recall, not in search quality.
Compare AI models on the same question
Instead of trusting one assistant, Smarty asks one flagship model from each provider the same question — first from memory, then grounded on live search results — and keeps the disagreement visible.
Run a comparisonHow the comparison works
2 — Search + LLMs
The same models are re-run grounded on real Google results gathered across several language locales. You see how each model reads the same evidence.
3 — Agreement and divergence
Claims shared by several providers are marked as agreement; claims made by one model alone, with no retrieved page behind them, are labelled unverified.
Which models are compared
Smarty uses latest-generation models only, and never two models from the same vendor family — a comparison is only useful across providers.
- GoogleGemini flagship, also used for the grounded Google search lane
- OpenAIGPT flagship, run on Azure OpenAI when configured
- AnthropicClaude flagship, run via AWS Bedrock when configured
- xAIGrok flagship
- MistralMistral flagship
- DeepSeekDeepSeek flagship
The roster tracks each provider's current flagship, so the exact model names change as providers ship new generations. Models without configured credentials are skipped rather than substituted with an older version.
Why compare at all
One model's confident paragraph hides what it does not know. Comparing providers on the same grounded evidence shows where they actually agree, where they contradict each other, and which claims no source supports.