Model registry - capabilities checked October 18, 2026
AI model API capabilities
Seven API features for each of the 167 models in the registry, each answer linked to the provider's own docs. Tool calling is the most common (150 models) and audio input the rarest (18).
Not stated means the provider's docs do not say either way. It is not the same as No.
Capabilities
- 150 of 167 modelsTool callingCheapest: GLM-4-32B-0414-128K, $0.10 per 1M output tokens.
- 97 of 167 modelsStructured outputsCheapest: Ministral 3 3B, $0.10 per 1M output tokens.
- 118 of 167 modelsImage inputCheapest: Ministral 3 3B, $0.10 per 1M output tokens.
- 18 of 167 modelsAudio inputCheapest: MiMo-V2.6-Flash, $0.28 per 1M output tokens.
- 136 of 167 modelsPrompt cachingCheapest: Ministral 3 3B, $0.10 per 1M output tokens.
- 97 of 167 modelsBatch APICheapest: Ministral 3 3B, $0.10 per 1M output tokens.
- 123 of 167 modelsReasoningCheapest: Qwen3.7-Flash, $0.13 per 1M output tokens.
Providers by capability
How many of each provider's models have each capability. Open a provider for the model-by-model breakdown, or a capability for every model that has it, sorted by price.
| Provider | Models | Tool calling | Structured outputs | Image input | Audio input | Prompt caching | Batch API | Reasoning |
|---|---|---|---|---|---|---|---|---|
| AI21 Labs | 2 | 2 | 0 | 0 | 0 | 0 | 0 | 0 |
| Alibaba Cloud (Qwen) | 27 | 25 | 5 | 19 | 3 | 17 | 15 | 21 |
| Amazon | 5 | 5 | 0 | 4 | 0 | 4 | 5 | 1 |
| Anthropic | 16 | 16 | 16 | 16 | 0 | 16 | 16 | 16 |
| Baidu | 6 | 0 | 0 | 3 | 0 | 4 | 0 | 1 |
| ByteDance Seed | 4 | 4 | 3 | 4 | 2 | 4 | 2 | 4 |
| Cohere | 5 | 4 | 3 | 0 | 0 | 0 | 0 | 0 |
| DeepSeek | 2 | 2 | 0 | 1 | 0 | 2 | 0 | 2 |
| 11 | 11 | 11 | 11 | 11 | 11 | 11 | 11 | |
| Inception | 3 | 2 | 2 | 0 | 0 | 3 | 0 | 2 |
| MiniMax | 5 | 5 | 0 | 1 | 0 | 5 | 0 | 5 |
| Mistral AI | 8 | 8 | 8 | 7 | 0 | 8 | 8 | 3 |
| Moonshot AI | 4 | 4 | 3 | 4 | 0 | 4 | 2 | 4 |
| OpenAI | 36 | 33 | 32 | 32 | 0 | 29 | 32 | 26 |
| StepFun | 5 | 4 | 4 | 3 | 0 | 5 | 0 | 4 |
| Tencent | 5 | 2 | 2 | 0 | 0 | 2 | 0 | 2 |
| xAI | 8 | 8 | 8 | 8 | 0 | 8 | 4 | 7 |
| Xiaomi MiMo | 2 | 2 | 0 | 2 | 2 | 2 | 2 | 2 |
| Z.ai | 13 | 13 | 0 | 3 | 0 | 12 | 0 | 12 |
| All providers | 167 | 150 | 97 | 118 | 18 | 136 | 97 | 123 |
What each capability means
- Tool calling
- The API lets the model call functions you define. You describe each tool with a JSON schema, the model replies with a structured call, and your code runs it.
- Structured outputs
- The API enforces a JSON schema you supply, so the response matches it. A plain JSON mode that only promises valid JSON does not count.
- Image input
- You can send images in the API request and the model reads them.
- Audio input
- You can send audio directly in the API request. A separate transcription step or speech-to-text model does not count.
- Prompt caching
- The provider offers prompt caching for the model, so a repeated prompt prefix can be reused instead of processed again from scratch.
- Batch API
- The model is available on the provider's batch API, which takes a set of requests and returns the results asynchronously.
- Reasoning
- The provider documents a thinking mode for the model: it can reason before it answers, either always on or switched on per request.
Newsletter
Get model price changes as they ship
One email per week. New model releases, price cuts and what they mean for your bill.