# SubQ Savant 1.4

> SubQ Savant 1.4 is Subquadratic's long-context reasoning model. It reads up to 4M tokens per request and reasons before it answers, with text, image and file input on OpenAI-compatible Chat Completions and Responses endpoints. It is in early access and enabled per organization.

**ID**: `subquadratic/subq-savant-1.4`  
**Creator**: subquadratic  
**Category**: text  
**Context**: 4.2M tokens  
**Input**: $0.10/M  
**Output**: $0.50/M  
**Released**: 2026-11-11  
**Web page**: https://anyrouter.dev/model/subquadratic/subq-savant-1.4

**Input modalities**: text, image, file  
**Output modalities**: text  

**Capabilities**: chat, reasoning, function-calling, structured-output, streaming

**Supported parameters**: max_tokens, temperature, top_p, stop, reasoning_effort, response_format, tools, tool_choice

**Aliases**: `subquadratic/subq-savant-latest`

## Usage

**Endpoint**: `POST https://anyrouter.dev/api/v1/chat/completions`

```bash
curl https://anyrouter.dev/api/v1/chat/completions \
  -H "Authorization: Bearer $ANYROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "subquadratic/subq-savant-1.4",
  "messages": [
    {
      "role": "user",
      "content": "Say hi in 3 words."
    }
  ]
}'
```

## Providers

| Provider | Input | Output |
| --- | --- | --- |
| subq | $0.10/M | $0.50/M |
| subq-byok | $0 | $0 |

## Model card

- Native context: 4194304 tokens [1]

### Strengths [1,2]

- Reads up to 4M tokens in a single request
- Reasons before answering with low, high and max effort levels
- Function calling, structured outputs and file input on OpenAI-compatible endpoints

### Known limits [1,2]

- Early access; requests return 403 until SubQ enables the model for the organization
- Always reasons and samples at temperature 1; turning reasoning off or changing temperature returns 400
- Image input only, with no image output

### Use cases [1]

- Whole-repository and long-document reasoning
- Contract, filing and data-room analysis over very large inputs
- Long-horizon agents that keep the full task in context

### API notes [2,1]

- Chat Completions and Responses endpoints; use an OpenAI SDK against https://api.subq.ai/v1
- reasoning_effort accepts low, high or max; none returns 400
- Reasoning tokens bill as output and count toward max_tokens
- Cached input bills at $0.01 per 1M tokens

### Links

- docs: https://docs.subq.ai/using-the-api/models/
- pricing: https://docs.subq.ai/using-the-api/models/
- Previous version: https://anyrouter.dev/model/subquadratic/subq-preview

### Sources

1. [SubQ models](https://docs.subq.ai/using-the-api/models/) (accessed 2026-11-18)
2. [SubQ Chat Completions API](https://docs.subq.ai/endpoints/chat-completions-api/) (accessed 2026-11-18)
3. [SubQ release notes](https://docs.subq.ai/using-the-api/release-notes/) (accessed 2026-11-18)

## Source

- Upstream docs: https://docs.subq.ai/using-the-api/models/
