Thinking Machines Releases Inkling 2 Mini
Understand the event
Hugging Face and Thinking Machines Lab released Inkling 2 Mini, a multimodal mixture-of-experts model with about 26B total parameters and 3B active per token, designed for local AI setups. It accepts text and images, supports a 1M-token context, and is available with support for Transformers, llama.cpp, MLX, and SGLang. A research variant allows one or two loop passes instead of three for faster inference, though generations may be slightly lower quality; the authors say they have not evaluated it in depth.
From Hugging Face Blog
Read next
More changes in today’s briefingHugging Face and Thinking Machines Lab Release Inkling 2 MiniDevelopments
5 developments
- 11/21 08:47 (UTC+8) · 1 reportsMira Murati Shared a Joke About Flying Anywhere to Benchmark Inkling 2 MiniMira Murati · X:The Inkling 2 Mini Team Asked to Fly Anywhere to Make a Plane Benchmark
- 11/21 08:14 (UTC+8) · 1 reportsThinking Machines Highlights Local Installation of Inkling 2 Mini Through OllamaThinking Machines · X:Thinking Machines Shares Ollama’s One-Command Setup for Inkling 2 Mini
- 11/21 08:04 (UTC+8) · 1 reportsGoodfire Studies Information Flow in Inkling 2 MiniThinking Machines · X:Thinking Machines Thanks Goodfire for Studying Looped Transformers
- 11/21 05:21 (UTC+8) · 1 reportsOllama Adds Inkling 2 Mini to Its Pre-ReleaseOllama · X:Ollama Adds Thinky Machines’ Inkling 2 Mini in Pre-Release
- 11/20 08:00 (UTC+8) · 7 reportsThinking Machines Releases Inkling 2 MiniHugging Face Blog:Hugging Face and Thinking Machines Lab Release Inkling 2 Mini
Report timeline
Follow the reports to explore different sides of the event.
- The Inkling 2 Mini Team Asked to Fly Anywhere to Make a Plane Benchmark
Yesterday the team asked for approval to fly “to anywhere” to make a plane bench for Inkling 2 Mini:
- Thinking Machines Shares Ollama’s One-Command Setup for Inkling 2 Mini
Thinking Machines says Ollama lets users install and run Inkling 2 Mini locally with a single command, and links to Ollama’s post with the setup details. Ollama says the model is available in version 0.40.17 pre-release and can run on a Mac with 32GB or more of memory.
- Thinking Machines Thanks Goodfire for Studying Looped Transformers
We're grateful for our research partner working with us to study looped transformers. Goodfire explored how layers pass information to future loops, and which interpretability tools allow us to understand the model's inner workings.
- Ollama Adds Thinky Machines’ Inkling 2 Mini in Pre-Release
Ollama has added Thinky Machines’ Inkling 2 Mini in version 0.40.17, available in pre-release. The 5.4B active-parameter model is described as small enough to run on a Mac with 32GB+ memory, and users can run it locally with `ollama run inkling-2-mini` or launch it in Pi.
- Thinking Machines Releases Inkling 2 Mini Under Apache 2.0
Thinking Machines is releasing Inkling 2 Mini under the Apache 2.0 license today. Clément Delangue says it reaches the Pareto frontier for open agent models and links to the model on Hugging Face.
- Mira Murati · XPicksInkling 2 Mini Is a Looped Transformer That Runs on a Laptop
Mira Murati says Inkling 2 Mini is a looped transformer that runs on a laptop. She says looping lets a model think more and write less, and that its thought process remains monitorable at this scale, though this may not remain true at larger scales. She also says a variable-loop research model is being shared for further study.
- Thinking Machines Makes Inkling 2 Mini Available on Tinker
Thinking Machines says Inkling 2 Mini runs on Macs with 32GB of RAM through Ollama and on NVIDIA's DGX Spark. The model is also available on Tinker for inference and fine-tuning, and in the Tinker Playground.
- Inkling 2 Mini is a looped transformer with 24 layers. The middle 12 layers loop three times; the first and last six run once. Looping lets a model reason more while writing less, which has raised concerns about monitorability. We studied this as part of our release.
Inkling 2 Mini is a looped transformer with 24 layers. The middle 12 layers loop three times; the first and last six run once. Looping lets a model reason more while writing less, which has raised concerns about monitorability. We studied this as part of our release.
- Thinking Machines Releases a Variable-Loop Research Model
Thinking Machines says its experiments found that no-CoT capability rises with more passes. At the tested scale, the chain of thought became shorter but remained legible, with no meaningful drop in monitorability. The company is releasing a variable-loop research model for others to run their own experiments.
- Thinking Machines Releases 26B-Parameter Inkling 2 Mini
Thinking Machines introduces Inkling 2 Mini, a 26B-parameter looped transformer it says is small enough to run on a laptop and powerful enough for agentic work. The company reports a score of 20.1 on AA Index, contrasting it with Inkling 2 Small at 594B parameters.
- Hugging Face BlogPicksHugging Face and Thinking Machines Lab Release Inkling 2 Mini
Hugging Face and Thinking Machines Lab released Inkling 2 Mini, a multimodal mixture-of-experts model with about 26B total parameters and 3B active per token, designed for local AI setups. It accepts text and images, supports a 1M-token context, and is available with support for Transformers, llama.cpp, MLX, and SGLang. A research variant allows one or two loop passes instead of three for faster inference, though generations may be slightly lower quality; the authors say they have not evaluated it in depth.
Keep up with AI, one email a day
Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue
Event heat trend
- Current within comparable coverage
- 33
- Peak within comparable coverage
- 3811/21 06:00 (UTC+8)
- 24-hour change
- –
Trends compare the same continuously observed participants, which may cover fewer sources than current heat. Point or click to inspect hourly heat, or use left and right arrow keys.