Cross-provider insight
Capability adoption by provider
What share of each provider's current lineup supports each capability. Greener cells mean broader adoption; gray means sparse or none.
| Provider | Active models | Vision | Function calling | Structured output | Prompt caching | Batch API | Fine-tuning |
|---|---|---|---|---|---|---|---|
| Google Gemini | 46 | 17% | 17% | 17% | 24% | 11% | 0% |
| OpenAI | 44 | 43% | 48% | 48% | 34% | 0% | 16% |
| Mistral AI | 42 | 14% | 2% | 0% | 0% | 0% | 0% |
| Cohere | 25 | 16% | 20% | 0% | 0% | 0% | 0% |
| AWS Bedrock | 14 | 36% | 36% | 0% | 0% | 7% | 14% |
| Anthropic | 13 | 85% | 92% | 0% | 77% | 0% | 0% |
| xAI | 9 | 33% | 44% | 33% | 33% | 0% | 0% |
| Azure OpenAI Service | 2 | 50% | 0% | 0% | 0% | 0% | 0% |
How this is built
Each cell is the percentage of a provider's currently available models that support that capability — based on the most recent record we have for each model. Preview and deprecated models still count (you can still use them); retired models don't. Hubs and aggregators that don't run their own model lineup aren't included.
Updated Jun 3, 2026 18:27 UTC · Full methodology →