Skip to content
TrustList
I2
AI Model

Inkling 2 Mini

Thinking Machines Lab's open-weights Inkling 2 Mini of 20 November 2026: Apache 2.0, 26.4 billion parameters with 3.4 billion active, text and image input, a context of up to one million tokens, and a 20GB checkpoint that runs on one workstation.

TrustList summary

Not yet independently verified. Thinking Machines Lab's own model card on Hugging Face (release date 20 November 2026, licence Apache 2.0) and a separate quantised copy published by Unsloth on Hugging Face on 20 November 2026 back these facts. The benchmark table on the model card is the vendor's own comparison, run by the vendor, so no scores are stored here. The card gives no price for API use on Tinker, so none is recorded. We will update this when it can be confirmed, and remove this note.

Thinking Machines Lab released Inkling 2 Mini on 20 November 2026, with open weights under the Apache 2.0 licence. Its larger sibling, Inkling 2 Small, came out on 5 November on the company's Tinker platform first. The Mini is the one a team can download and run itself.

Small enough to run locally

The model has 26.4 billion parameters, of which 3.4 billion are active for each token. It ships as a single checkpoint of about 20GB in mixed NVFP4 and MXFP8 precision. The company lists one NVIDIA DGX Spark, or an Apple silicon machine with 32GB or more of memory, as enough for inference, and recommends 32GB of video memory on other GPUs. It names support in vLLM, SGLang, llama.cpp, Ollama and other runtimes, and Unsloth published a quantised GGUF copy on the same day.

A looped design

The card describes a 24-layer mixture-of-experts transformer in which the middle 12 layers run three times with shared weights, so each token passes through 48 layer applications. Each token is routed to 8 of 128 experts. The company says each extra pass adds about one billion effective active parameters, giving about 5.4 billion in all. It is the first looped transformer the company has released, and the card says it studied whether the chain of thought can still be monitored.

What it is for

The company aims the model at agents and tool use, coding assistants, chat and retrieval systems. It reads text and images and writes text, with a context window of up to one million tokens. It can also be used and fine-tuned through Tinker.

What the company reports

The model card compares the Mini with four other open models of similar size on coding, agent, vision and reasoning tests, and reports internal safety tests. These are the company's own runs, and they are not stored as scores here.

Key facts

Maker: Thinking Machines Lab Released: 20 November 2026 Parameters: 26.4B total, 3.4B active Inputs: text and images Context window: up to 1,000,000 tokens Licence: Apache 2.0, open weights Checkpoint: about 20GB (NVFP4 and MXFP8)

Benchmarks & AI stats

BenchmarkOfficialCommunity avg
No benchmark scores yet. Be the first to add one.

“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).

Sign in to add a benchmark score for this model.

Inkling 2 Mini launches

Request a demo or quote from Inkling 2 Mini

Protected by reCAPTCHA - Google Privacy Policy and Terms apply.

By sending, you agree we may share your request and contact details with the provider once you confirm your email.

Reviews

Write the first review of Inkling 2 Mini

Used it? Your experience helps other buyers decide.

Write a review

Questions & answers

No questions yet. Be the first to ask about Inkling 2 Mini.

Inkling 2 Mini alternatives

Similar listings buyers compare against this one.

Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems

Black Forest Labs' image generation and editing model with bounding-box layouts, multiple reference images and 4K output

Google's open multimodal embedding model that maps text, images, audio and video into one vector space, under Apache 2.0

Anthropic's smallest and fastest Claude 5.5 model, available on its own API and three cloud platforms

Cantina Security's first open-weight security model, a post-trained GLM-5.3-Flash for vulnerability research and exploit checks

JetBrains' open 12B mixture-of-experts coding model with 2.5B parameters active per token under Apache 2.0

Open-weight translation model from Bilibili's Index team, covering 150 languages and following glossary and style instructions

Microsoft AI's high-volume, low-latency variant of MAI-Voice-2.1 for speech in 23 languages, aimed at voice agents