LifeHubber.
Resources
AI Resources
Inkling 2 Small
Inkling 2 Small is an open-weight model from Thinking Machines Lab. It takes text and images and replies in text.
You can try it in the Tinker Playground. Running the published weights yourself takes hundreds of gigabytes of GPU memory.
What it is
Text answers about images and code
Send a screenshot or diagram alongside a written question. The model can return a description, an explanation, or code as text.
Why it stands out
A fraction of the model works on each token
The model has 594 billion parameters. Its mixture-of-experts design activates about 17 billion for each token, while a deployment still has to make the full set of weights available.
Availability
Two downloadable formats and hosted access
Thinking Machines publishes NVFP4 and MXFP8 weights on Hugging Face and lists Apache 2.0 on the model card. Tinker and third-party providers offer hosted access.
Why it matters
What makes it useful
If a web layout looks wrong, you could share a screenshot with the relevant code and ask which part of the layout to inspect. The screenshot shows the symptom and the code gives the model something specific to reason about.
What to know
Where it fits
The Playground is a way to try a prompt before integrating the model. For a customized version, Thinking Machines documents fine-tuning through Tinker. Teams running the published weights can find setup guides under Getting Started on Hugging Face.
Notable points
What stands out
Thinking Machines trained the model using NVFP4 and recommends that format. The card lists a minimum of 500 GB of combined GPU memory for NVFP4 and 800 GB for MXFP8. Those figures are for serving the model. The main Hugging Face repository lists about 394 GB of files on disk.
Before using
What to review
The model card lists up to one million tokens of context. Check the limit on the service or configuration you plan to use.
Check the selected provider’s data handling before sending screenshots or documents that contain private information.
The model card notes that answers can include unsupported details and that performance can degrade in long conversations.
Check the original listing's documentation for current requirements, limitations, licensing and any relevant privacy or security considerations.
Reader fit
Who may find it relevant
This is a fit for developers testing a text-and-image workflow, or infrastructure teams comparing which weight format their GPU cluster can serve. The documented inputs are text and images, so audio needs a separate path.
Editorial note
Why LifeHubber lists it
Inkling 2 Small is worth listing for teams that want to try visual troubleshooting before deciding where to run it. You can start in the Playground with a screenshot that contains no private information. If the model is useful and you need to keep later screenshots on infrastructure you control, the published weights give you a route to assess, with GPU memory requirements you can check before committing to it.
Source links
Source materials
Reader note
About our listings
LifeHubber’s listings help you explore what’s available in AI and are not endorsements. We don’t independently verify every detail or guarantee that each listing is accurate, safe, actively maintained, or right for your needs.
What to explore next
Compare the inputs before picking a checkpoint
If your workflow also needs to interpret recorded speech, look at the original Inkling model.
More in AI Models
Keep browsing this category
Explore more AI model resources.
Gemma 4
google/gemma-4
A Google DeepMind Gemma 4 model family collection with public checkpoints including Gemma 4 12B, a dense multimodal model Google describes around local agentic workflows, native audio input, and encoder-free vision/audio handling.
MiniMax-H3-Image
MiniMaxAI/MiniMax-H3-Image
MiniMax’s still-image generation and editing model with public weights, up to 16 reference images, roughly 16-megapixel output, and documented hosted and local workflows with a separate prompt-expansion stage.
TRELLIS.2
microsoft/TRELLIS.2
A Microsoft image-to-3D model with O-Voxel geometry, PBR materials, GLB export, inference examples and training code.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
MORE ON LIFEHUBBER
Use AI on your own terms.
Explore more AI projects, stories, guides, tools and community insights across LifeHubber.
