# Trinity Large Thinking

> Trinity Large Thinking is Arcee AI's open-weights reasoning model, a 398B sparse mixture-of-experts with about 13B active parameters per token. It is post-trained with extended chain-of-thought and agentic reinforcement learning, and is built for tool calling, multi-step planning and agent workflows. It writes its reasoning in think blocks before the answer.

**ID**: `arcee-ai/trinity-large-thinking`  
**Creator**: arcee-ai  
**Category**: text  
**Context**: 262K tokens  
**Pricing**: $0.25 / $0.80 per 1M · billed by your provider (BYOK)  
**Released**: 2026-04-01  
**Web page**: https://anyrouter.dev/model/arcee-ai/trinity-large-thinking

**Input modalities**: text  
**Output modalities**: text  

**Capabilities**: chat, streaming, function-calling, reasoning, agentic

**Supported parameters**: max_tokens, temperature, top_p, top_k, tools, tool_choice, reasoning, include_reasoning

## 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": "arcee-ai/trinity-large-thinking",
  "messages": [
    {
      "role": "user",
      "content": "Say hi in 3 words."
    }
  ]
}'
```

## Providers

| Provider | Input | Output |
| --- | --- | --- |
| openrouter-byok | $0.25/M (list) | $0.80/M (list) |

BYOK routes are billed by your own provider key — AnyRouter charges $0. Rates marked (list) are the provider's published list price.

## Model card

- Architecture: mixture of experts [1]
- Total parameters: 398B [1]
- Active parameters: 13B [1]
- License: OpenMDW-1.1, open weights (https://huggingface.co/arcee-ai/Trinity-Large-Thinking) [1]

### Strengths [1]

- Built for tool calling, multi-step planning and agent workflows
- Reports 94.7% on Tau2-Telecom and 98.2% on LiveCodeBench

### Known limits [1]

- Reasoning tokens must stay in the history for multi-turn and agent loops

### Use cases [1]

- Agent and tool-calling backbones
- Long-chain reasoning tasks

### API notes [2]

- Card recommends temperature 0.3
- Reasoning is returned in think blocks before the final answer

### Links

- model card: https://huggingface.co/arcee-ai/Trinity-Large-Thinking
- paper: https://arxiv.org/abs/2602.17004
- docs: https://docs.arcee.ai/arcee-language-models/trinity-large-thinking.md

### Sources

1. [arcee-ai/Trinity-Large-Thinking model card](https://huggingface.co/arcee-ai/Trinity-Large-Thinking/raw/main/README.md) (accessed 2026-10-17)
2. [Arcee docs, Trinity-Large-Thinking](https://docs.arcee.ai/arcee-language-models/trinity-large-thinking.md) (accessed 2026-10-17)

## Source

- Upstream docs: https://docs.arcee.ai/arcee-language-models/trinity-large-thinking.md