Skip to main content

Thinking Machines LabLanguage modelResearch preview

TML-d24-varloop-research

Summary

TML-d24-varloop-research is a separate research checkpoint from Thinking Machines Lab with a variable number of passes through shared layers. Choosing between one and five passes changes the compute required per token. Its open weights are a research preview, not intended for production, and have no API access.

All specifications

Architecture

Checkpoint
thinkingmachines/TML-d24-varloop-research
Source: Checkpoint
Model class
Multimodal autoregressive looped transformer for research.
Source: Model class
Parameters
26.4 billion total and 3.4 billion active. Effective compute varies with the number of loop passes.
Source: Parameters

Input and output

Inputs and outputs
Text and images in. Text out.
Source: Inputs and outputs
Context window
Up to 1M tokens
Source: Context window

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking
Unknown
no source
Watermarking technique
Not known
no source

Licence and access

Access
Open weights. No API access, including through Tinker.
Source: Access
License
Apache 2.0
Source: License
Weight format
Mixed NVFP4, MXFP8, and BF16 precision, about 20 GB.
Source: Weight format
Shutdown
None announced
no source

Further details

Distinction from Inkling 2 Mini
A separate research checkpoint. Later training stages differ from Inkling 2 Mini. The provider cautions against direct performance comparisons.
Source: Distinction from Inkling 2 Mini

Frequently asked questions about TML-d24-varloop-research

How large is the context window of TML-d24-varloop-research?

1 M tokens. Separate limits may apply to input and output.

Context windowContext limit in tokens · higher is better
TML-d24-varloop-researchRank 40 of 1161M
Claude Fable 5.21M
Claude Opus 5.51M
Claude Fable 5.11M
Claude Sonnet 5.51M
Muse Wonder Full1M
Claude Opus 51M

Can I run TML-d24-varloop-research locally?

Yes, the weights are open.

Stated by Thinking Machines Lab

Open weights
yes
License
Apache 2.0
Parameters
26.4B
active per token
3.4B