# Inkling 2 Mini

> Inkling 2 Mini is Thinking Machines Lab's open-weights multimodal model. It is a 26.4B-parameter mixture-of-experts looped transformer with 3.4B active parameters, a 1M-token context and image input. Trained in NVFP4, it runs in about 20GB.

**ID**: `thinkingmachines/inkling-2-mini`  
**Creator**: thinkingmachines  
**Category**: multimodal  
**Context**: 262K tokens  
**Pricing**: $0.10 / $0.40 per 1M · billed by your provider (BYOK)  
**Released**: 2026-11-20  
**Links**: [Weights on Hugging Face](https://huggingface.co/thinkingmachines/Inkling-2-Mini) · [Announcement](https://thinkingmachines.ai/news/inkling-2-mini/) · [AnyRouter release](https://anyrouter.dev/blog/releases/2026-11-21-inkling-2-mini)  
**Web page**: https://anyrouter.dev/model/thinkingmachines/inkling-2-mini

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

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

## 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": "thinkingmachines/inkling-2-mini",
  "messages": [
    {
      "role": "user",
      "content": "Say hi in 3 words."
    }
  ]
}'
```

## Providers

| Provider | Input | Output |
| --- | --- | --- |
| thinkingmachines-byok | $0.15/M (list) | $0.50/M (list) |
| openrouter-byok | $0.10/M (list) | $0.40/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 [1,4]
- Total parameters: 26.4B [1,4]
- Active parameters: 3.4B [1,5]
- Native context: 1048576 tokens [4,5]
- License: Apache-2.0, open weights (https://huggingface.co/thinkingmachines/Inkling-2-Mini) [3,1]

### Strengths [1]

- Leads every open-weights model under 100B total parameters on agentic coding (DeepSWE v1.1 48.2%) and the Vals Index v2.2 (46.5%).
- Looped transformer: the middle layers run three times per token, adding serial reasoning without adding parameters.
- Runs natively in NVFP4 in about 20GB, small enough for a consumer laptop, and is fast for its quality.

### Known limits [1]

- Not the strongest small model for multi-turn chat or instructions with many constraints: 50.0% on MultiChallenge and 33.4% on ComplexConstraints v1.1.
- Trails larger models on reasoning-intensive benchmarks such as CritPt v2 and HLE-Diamond.
- Looping does not add stored knowledge, so factuality (AA-Omniscience) is in line with other same-size models; it relies on search and tools instead.

### Use cases [1]

- Long-horizon agentic coding and autoresearch on local hardware.
- Fast local or offline inference on a laptop, including multiple persistent agents on one machine.
- Search- and tool-heavy knowledge work, and research into monitorability of looped transformers.

### API notes [6,1,4]

- First-party serving via Tinker's OpenAI-compatible /chat/completions endpoint (https://tinker.thinkingmachines.dev/services/tinker-prod/oai/api/v1).
- Reasoning effort is a float from 0.0 to 0.99 (chat template maps none/minimal/low/medium/high/xhigh/max).
- Image input is accepted on the inference API; audio input is not supported.

### Links

- announcement: https://thinkingmachines.ai/news/inkling-2-mini/
- model card: https://thinkingmachines.ai/model-card/inkling-2-mini/
- docs: https://tinker-docs.thinkingmachines.ai/inference/openai/overview/
- pricing: https://tinker-docs.thinkingmachines.ai/tinker/models/models_and_pricing/
- Previous version: https://anyrouter.dev/model/thinkingmachines/inkling-2-small

### Sources

1. [Introducing Inkling 2 Mini - Thinking Machines Lab](https://thinkingmachines.ai/news/inkling-2-mini/) (accessed 2026-11-21)
2. [Inkling 2 Mini Model Card - Thinking Machines Lab](https://thinkingmachines.ai/model-card/inkling-2-mini/) (accessed 2026-11-21)
3. [Inkling 2 Mini released weights and Apache-2.0 license](https://huggingface.co/thinkingmachines/Inkling-2-Mini) (accessed 2026-11-21)
4. [Inkling 2 Mini model configuration (1,048,576 context, loop config)](https://huggingface.co/thinkingmachines/Inkling-2-Mini/raw/main/config.json) (accessed 2026-11-21)
5. [Tinker Docs - Models & Pricing](https://tinker-docs.thinkingmachines.ai/tinker/models/models_and_pricing/) (accessed 2026-11-21)
6. [Tinker Docs - OpenAI API Compatible Inference](https://tinker-docs.thinkingmachines.ai/inference/openai/overview/) (accessed 2026-11-21)

## Source

- Upstream docs: https://thinkingmachines.ai/news/inkling-2-mini/
