Thinking Machines
Released November 20, 2026

Inkling 2 Mini

Inkling 2 Mini by Thinking Machines is a looped transformer MoE that runs on laptops, reusing middle layers three times per token for strong agentic coding.

Inputs
Text
Image
Outputs
Text

Model Overview

Capabilities, design details, and architectural traits

Inkling 2 Mini - a looped transformer small enough for a laptop

Inkling 2 Mini is Thinking Machines' open-weights model built to run on consumer laptops and desktop AI computers. Its defining idea is looping: the middle layers of the model run three times for every token, letting it apply more reasoning and reach the right answer with fewer tokens, which helps most on reasoning-heavy work like agentic coding. It is available on Tinker for inference and customization.

TraitDetail
Looped architectureMiddle layers execute three times per token, adding reasoning depth without adding parameters
Laptop-class footprintDemonstrated running offline on a MacBook Pro in airplane mode, where it built a playable flight simulator in minutes
Native low precisionMost weights trained and deployed natively in NVFP4, with the last four layers in MXFP8
Research companionReleased alongside a variable-loop variant for studying how loop depth affects performance and monitorability

Reasoning without writing it down

Looping lets the model reason longer without writing its reasoning into its chain of thought. Thinking Machines frames this through monitorability, the ability to oversee a model by reading its CoT. In tests with the variable-loop variant, the no-CoT time horizon improved from 16 seconds at one pass to 37 seconds at three, serial-reasoning performance kept improving out to ten loops (twice the trained depth), and the chain of thought grew shorter with more loops but stayed legible. Scores on tests of agentic sabotage did not trend higher as loops were added.

Local-first design

The model's strengths mirror those of the larger Inkling 2 Small: it does best on long-horizon agentic work and is the best among open-weights models of similar size on most measured tasks. It closes knowledge gaps by efficiently using search and tools rather than storing more facts.

Benchmark Performance

Independent evaluations · Artificial Analysis

20.1%
Intelligence

Compare Models Side-by-Side

Evaluate specifications, pricing, and independent benchmark indices

Model Details
General Info
ProviderThinking MachinesAnthropicOpenAI
Release DateNovember 20, 2026November 9, 2026November 17, 2026
Knowledge Cutoff---
Context & Limits
Context Window1M1M
1.1M
Best Context Window
Pricing (per 1M tokens)
Input Pricing
$0.15
Best Input Pricing
$8$8
Output Pricing
$0.50
Best Output Pricing
$40$40
Modalities
Inputs
textimage
textimagefile
textimagefile
Outputs
text
text
text
Benchmarks (0-100)
Intelligence Index20.1
64.6
Best Intelligence Index
62.1
Coding Index---
Agentic Index---

Benchmark comparison not available

Select at least two models that have available benchmark evaluation data to render the performance comparison charts.

Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.

Explore more from Thinking Machines

Other models by Thinking Machines

Top AI Models

Leading alternatives by intelligence score

View all