Inkling 2 Mini at a glance
Thinking Machines Lab released Inkling 2 Mini on November 20, 2026: an open-weights model (Apache 2.0) with 26.4B parameters in a roughly 20GB file that runs on a laptop. In the lab's reported results of November 20, 2026, it leads every open-weights model under 100B parameters on DeepSWE v1.1, at 48.2%.
Current standing
As of
- Rank
- #42of 51 models
Inkling 2 Mini · 20.1 on the Artificial Analysis Intelligence Index v5.0.1 (opens in a new tab)
- Recorded at publication
- 20.1 (v5.0.1)
as of
- Current leader
- Claude Fable 5.2 (Anthropic) · 64.6
Computed from our daily snapshot; it changes as new models are measured. Thinking Machines Lab releases Inkling 2 Mini, an open-weights looped-transformer model that runs on a laptop
Ask your AI about this story
Ask what changed in Inkling 2 Mini and what to check before using it in your business. Compare it on price, speed, intelligence and the price-to-speed ratio against the models I would otherwise use, and say which tasks it suits or does not suit.
What changed with Inkling 2 Mini
Thinking Machines Lab released Inkling 2 Mini on November 20, 2026, an open-weights multimodal model small enough to run on a laptop. Inkling 2 Mini is a Mixture-of-Experts model with 26.4 billion total parameters and 3.4 billion active per token; its middle 12 layers run three times per token (a looped transformer), giving 5.4 billion effective active parameters. It accepts text and images, writes text, and its model card lists a context window of up to 1 million tokens. The weights are a single checkpoint of about 20GB in NVFP4 and MXFP8 precision, published on Hugging Face under the Apache 2.0 license, and the model also runs on Tinker, Thinking Machines' platform, for inference and fine-tuning.
In Thinking Machines Lab's reported results of November 20, 2026, Inkling 2 Mini leads every open-weights model with fewer than 100 billion total parameters on DeepSWE v1.1 (48.2% of tasks resolved) and the Vals Index v2.2 (46.5%), and it scores 77.9% on SWE-bench Pro v2 (Hard) and 79.2% on SWE-bench Verified in the same vendor table. Artificial Analysis scored Inkling 2 Mini at 20.1 on its Intelligence Index v5.0.1 as of November 20, 2026, and measured an output speed of 279.9 tokens per second, at a cost of $0.59 per index task. The lab also lists limits: 50.0% on MultiChallenge and 33.4% on ComplexConstraints v1.1, behind Qwen3.8-27B on both.
On Tinker, sampling costs $0.15 per million input tokens ($0.015 cached) and $0.50 per million output tokens, and fine-tuning costs $0.75 per million tokens on the 64K-context version and $1.50 on the 256K version, according to the Tinker pricing documentation. The open weights cost nothing to download: at launch the model runs on any Mac with at least 32GB of memory through Ollama, on NVIDIA's DGX Spark, and through SGLang, vLLM and GGUF builds from Unsloth and llama.cpp. Stanford researchers given early access measured 0.703 on their intelligence-per-joule test on a DGX Spark, ahead of Qwen3.6-35B-A3B at 0.430.
Inkling 2 Mini benchmarks and pricing
Values are copied from the linked source on the date shown. They are not IA Compa measurements.
- Artificial Analysis Intelligence Indexas of · Source for Artificial Analysis Intelligence Index: artificialanalysis.ai (opens in a new tab)
- 20.1 (v5.0.1)
- Output speed (Artificial Analysis)as of · Source for Output speed (Artificial Analysis): artificialanalysis.ai (opens in a new tab)
- 279.9 tokens/s
- Input price per 1M tokens (Tinker)as of · Source for Input price per 1M tokens (Tinker): tinker-docs.thinkingmachines.ai (opens in a new tab)
- $0.15
- Output price per 1M tokens (Tinker)as of · Source for Output price per 1M tokens (Tinker): tinker-docs.thinkingmachines.ai (opens in a new tab)
- $0.50
- Context window (model card)as of · Source for Context window (model card): thinkingmachines.ai (opens in a new tab)
- Up to 1M tokens
- DeepSWE v1.1 tasks resolved (Thinking Machines Lab)as of · Source for DeepSWE v1.1 tasks resolved (Thinking Machines Lab): thinkingmachines.ai (opens in a new tab)
- 48.2%
How it compares
Current models we have covered: released in the last six months, newest version of each line only. Each value is the model's latest recorded value. (Models)
Where to compare Inkling 2 Mini
Independent leaderboards and official resources for comparing this model yourself. Rankings change often; each site shows its own current data.
General leaderboards
Why Inkling 2 Mini matters for businesses
Why it matters: a capable open-weights model that runs on a laptop gives a small business private AI with no per-token bill and no data leaving the building: client files, contracts and support history can be processed offline on a machine the business owns. Agencies can build on-device features and persistent local agents for clients, fine-tune the model under Apache 2.0 terms, and reserve cloud models for the work that truly needs them, cutting inference spend on routine tasks.
Put it to work: AI services
Get AI News by email
Verified AI stories, each with its official source, on Monday and Thursday mornings. Add breaking news to hear about the biggest ones as soon as we publish them.
Every story links its official source
Free, in English or Spanish
Unsubscribe with one click, any time
Sources for this story
- Introducing Inkling 2 Mini (opens in a new tab)Official
Thinking Machines Lab · thinkingmachines.ai
- Inkling 2 Mini Model Card (opens in a new tab)Official
Thinking Machines Lab · thinkingmachines.ai
- Models & Pricing (Tinker documentation) (opens in a new tab)Official
Thinking Machines Lab · tinker-docs.thinkingmachines.ai
- Inkling 2 Mini weights on Hugging Face (Apache 2.0) (opens in a new tab)Official
Thinking Machines Lab · huggingface.co
- Inkling 2 Mini: intelligence, performance and price analysis (opens in a new tab)Benchmark
Artificial Analysis · artificialanalysis.ai
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #llm
- #thinking-machines
- #inkling
- #multimodal
- #mixture-of-experts
- #open-weights
- #looped-transformer
- #fine-tuning
