USASI
Model release

Inkling-Small

Release in the Inkling family · version Inkling-Small

Maintained by Thinking Machines Lab13

Inkling-Small is a 42-layer decoder-only mixture-of-experts transformer with 276B total and 12B active parameters, routing each token to 6 of 256 experts plus 2 shared experts. Like Inkling, it accepts text, image, and audio input, outputs text, and supports a context window of up to 1M tokens.12

Last reviewedEntry updated Documented release Jul 30, 2026

Availability and license

Overall availability

Public

BF16 and NVFP4 checkpoints are downloadable from Hugging Face without gating; the developer's Model Acceptable Use Policy also applies.34

Availability is separate from permission: read the license before using or redistributing.

The Hugging Face repository declares Apache 2.0 in its metadata and links to the Apache license text; it has no separate LICENSE file. The developer's Model Acceptable Use Policy states that anyone who accesses, downloads, or uses its model materials agrees to be bound by it, and it lists prohibited uses.34

Model-disclosure tier

Computed from the checklist below using USASI rubric v0.1. An editorial category, not a certification.
Model-disclosure tier (USASI rubric v0.1): Open-weight

The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.

How tiers are computed

Public materials checklist

Items for a model under USASI rubric v0.1. Unknown means unassessed or insufficient evidence.
Public materials checklist for Inkling-Small
ItemStatusNotes and evidence
WeightsCan the general public download the model parameters for this release?PublicPublished as a BF16 checkpoint and a separate NVFP4-quantized checkpoint.31
Inference codeIs code for running the model published?PublicThe model card documents deployment with open-source SGLang, vLLM, TokenSpeed, Unsloth, and Hugging Face Transformers.13
Training codeIs the code used to train the model published?UnknownNot assessed.
Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access.PartialThe model card and training-data documentation describe source categories (public, third-party, and synthetic data), deduplication, and filtering; the release post says the pretraining data mix differs from Inkling's. The data is not released.152
Training recipeAre the training configuration and procedure documented in enough detail to follow?PartialThe release post describes, at a high level, changes to the pre-training data mix and recipe, post-training that used on-policy distillation from Inkling in part, and further agentic-coding reinforcement learning.2
Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only.PartialBenchmark results are published in the model card and as evaluation result files in the Hugging Face repository.3

What it is useful for

The model card describes use by developers building agentic and tool-use systems, coding assistants, chatbots, and retrieval-augmented generation systems.1

Run and use notes

Documented facts only. No hardware or performance claims are made without a cited source and stated assumptions.
  • The model card states that the BF16 checkpoint needs at least 600 GB of aggregated GPU memory (for example 4x NVIDIA B300 or 8x H200), and that the NVFP4-quantized checkpoint needs at least 180 GB (W4A4 on 1x B300, which requires SM100+ architecture, or W4A16 on 2x H200).1

Organization context

Provenance and derivatives

Thinking Machines Lab says Inkling-Small was pretrained with a changed data mix and recipe, and that an earlier checkpoint was post-trained in part using on-policy distillation with Inkling as the teacher, followed by further agentic-coding reinforcement learning.2

  • Derived from: Inkling — Teacher model for on-policy distillation during post-training; not a weight initialization.

Other releases in the Inkling family

  • InklingModel-disclosure tier (USASI rubric v0.1): Open-weight

Inkling family overview

U.S. eligibility

Project eligibility rests on documented governing or maintaining entities, not on contributors.

Eligible · basis: U.S. headquarters

The model card names Thinking Machines Lab, Inc. as the model provider; the company states in its privacy notice that it is based in the United States (see the organization record for the San Francisco headquarters assessment).16

Assessed Sep 29, 2026

Sources

  1. 1.
    Inkling-Small Model Card (external site: thinkingmachines.ai)

    Thinking Machines Lab · Model card · published Jul 30, 2026 · accessed Sep 29, 2026

  2. 2.
    Introducing Inkling-Small (external site: thinkingmachines.ai)

    Thinking Machines Lab · Announcement · published Jul 30, 2026 · accessed Sep 29, 2026

  3. 3.
    thinkingmachines/Inkling-Small (external site: huggingface.co)

    Thinking Machines Lab (Hugging Face) · Model card · accessed Sep 29, 2026

  4. 4.
    Model Acceptable Use Policy - Thinking Machines Lab (external site: thinkingmachines.ai)

    Thinking Machines Lab · License · published Jul 15, 2026 · accessed Sep 29, 2026

  5. 5.
    Training Data Documentation - Thinking Machines Lab (external site: thinkingmachines.ai)

    Thinking Machines Lab · Documentation · accessed Sep 29, 2026

  6. 6.
    Privacy Notice - Thinking Machines Lab (external site: thinkingmachines.ai)

    Thinking Machines Lab · Official page · published May 19, 2026 · accessed Sep 29, 2026

This listing is not an endorsement, a safety assessment, or a federal approval.

Support Us

Help keep USASI useful.

Optional. No USASI account required. Payment takes place on the linked provider’s website (Buy Me a Coffee).

About supporting this project