What Thinking Machines Lab released
On 20 November 2026, Thinking Machines Lab published Introducing Inkling 2 Mini. The post says the company is releasing a fast, powerful model that is small enough to run on a laptop. Inkling 2 Mini is a looped transformer: it runs the middle layers of the model three times for every token. The official post and Hugging Face card describe a mixture-of-experts model with 26.4B total parameters, 3.4B active, and 5.4B weight activations per token to account for looping.
The official announcement says Inkling 2 Mini leads every open-weights model with less than 100B total parameters on DeepSWE v1.1 (48.2%) and Vals Index v2.2 (46.5%). Those figures are the company's published results, not independent measurements in this article. Most weights were trained and are deployed in NVFP4, with the last four layers in MXFP8. The official memory footprint is about 20GB. The post describes a demo on a MacBook Pro (M5 Pro, 48GB) in airplane mode, in which the model built a playable flight simulator offline.
Where to get it, the license, and official prices
The official card on Hugging Face, checked on 21 November 2026, states the license as Apache 2.0. The README metadata uses license apache-2.0 and links to the official Apache 2.0 license text. The Hugging Face tag on the same page also reads apache-2.0. The weights are on Hugging Face as thinkingmachines/Inkling-2-Mini. The announcement also releases a research-specific variable-loop variant at thinkingmachines/TML-d24-varloop-research.
The post says Inkling 2 Mini is available on Tinker today for inference and customization, and the official card adds Tinker Playground for chat. The live Tinker pricing page, labeled as of 19 October 2026 and checked on 21 November 2026, lists Inkling 2 Mini training and sampling rates, in USD per million tokens: at 64K context, prefill $0.15, cached prefill $0.015, sample $0.50, and train $0.75; at 256K, the same prefill, cached, and sample rates, with train $1.50. The same page's serverless inference table, also as of 19 October 2026, lists Inkling 2 Small, Inkling, and Inkling-Small, but not Inkling 2 Mini, so this article does not invent a serverless rate for Mini. Storage is $0.10 per GB per month.
Official local runtimes, context length, and limits
The official Hugging Face model card, which answers common search questions about Hugging Face, Ollama, Unsloth, GGUF, size, and benchmarks, says Inkling 2 Mini supports a context window of up to 1 million tokens, and accepts text and image inputs with text output. Official local inference support includes Hugging Face, SGLang, vLLM, TokenSpeed, NVIDIA Dynamo, Unsloth including quantized GGUF versions, llama.cpp for quantized GGUF versions, Llama App, and Ollama and Exo Labs on Apple silicon. The card recommends Apple silicon devices with 32GB or more memory, or other GPUs with 32GB or more of VRAM.
Official limitations include weaker factual recall than Inkling 2 Small, a risk of getting stuck in long reasoning loops, less reliable tool use than larger models, and the possibility of destructive file actions without confirmation. The card also says support for looped models in local runtimes is new, and that speed varies by runtime and hardware.
Go to the source
Sources & further reading
Based on the official documentation below. Features and access can change; check the provider before making a decision.
- Thinking Machines Lab: Introducing Inkling 2 Mini, dated 20 November 2026, rendered 21 November 2026 (opens in a new tab)
- Official Hugging Face model card for thinkingmachines/Inkling-2-Mini, including README license metadata, rendered 21 November 2026 (opens in a new tab)
- Official Tinker pricing page, labeled as of 19 October 2026, rendered 21 November 2026 (opens in a new tab)
Was this page useful?
Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.
Go a little further