Inkling 2 Small
Inkling 2 Small has 594 billion total and 17 billion active parameters. It takes text and image inputs and generates text.
What changed in this generation
The forty-eight-layer decoder routes each token to eight of 384 experts. Images pass through a vision transformer. The model card describes up to one million tokens of context and lists text and image inputs, not audio.Thinking Machines: Inkling 2 Small model card
Coding and visual feedback
Training includes long coding tasks, multiple agent environments and graders that examine the approach as well as the result. The model can use visual feedback while developing software. Its visual demonstrations use code and tools; it does not natively output image pixels.Thinking Machines: Inkling 2 Small release
Weight release remains pending
As of 6 November 2026, Tinker provides fine-tuning, sampling and Playground access. The company plans an Apache 2.0 weight release after external safety work. That plan does not yet provide downloadable weights or exported checkpoints.Thinking Machines: Inkling 2 Small release
How to use it well
Practical guidance. Use screenshots and documents as evidence for coding and analysis tasks. Verify that generated files and tool actions meet the request. Confirm the service context limit before relying on the larger model-card maximum.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API, Provider product |
| Model identifiers | thinkingmachines/Inkling-2-Smallthinkingmachines/Inkling-2-Small:peft:262144:sampling-nvfp4 |
| Context window | 262,144 tokens Tinker serverless inference beta; model-card maximum is larger Tinker: Current models and pricing |
| Input price | 0.3 USD per million tokens Tinker serverless beta, uncached input Tinker: Current models and pricing |
| Output price | 1 USD per million tokens Tinker serverless beta output Tinker: Current models and pricing |
| Current terms | Tinker Terms of Use apply; public weights are not yet releasedThinking Machines: Inkling 2 Small model cardThinking Machines: Inkling 2 Small release |
| API access | Serverless inference is beta; production use has a waitlistTinker: Current models and pricing |
Related models
- Inkling
Thinking Machines Lab · Inkling
A large open-weight model that reads text, images and audio.
Launch date not confirmedAPI · Downloadable · Provider product - Inkling-Small
Thinking Machines Lab · Inkling
A smaller multimodal Inkling model with public weights.
Launch date not confirmedAPI · Downloadable · Provider product