# Inkling

> **TL;DR.** Inkling is Thinking Machines Lab's first open-weights large language model: a 975-billion-parameter Mixture-of-Experts system with 41B active parameters that natively reasons over text, images, and audio.

- **Category:** AI / Foundation Models / Open Weights
- **Stage:** nascent
- **Age:** 1 days
- **Origin date:** 2026-07-15
- **First detected:** 2026-07-16
- **Canonical URL:** https://earlyterms.com/term/inkling
- **Sources:** 6 primary URLs

## Definition

Inkling is Thinking Machines Lab's first open-weights large language model: a 975-billion-parameter Mixture-of-Experts system with 41B active parameters that natively reasons over text, images, and audio. It is distinct from the unrelated 2010s Inkling ed-tech and drawing-tablet products of the same name.

Released [July 15, 2026](https://thinkingmachines.ai/news/introducing-inkling/) by Mira Murati's Thinking Machines Lab under Apache 2.0, Inkling trained on 45 trillion tokens and ships pre-wired for fine-tuning on the lab's Tinker platform. It scored 1,116 points on Hacker News in a day, with benchmarks trailing DeepSeek V4 Pro and GLM 5.2.

## Analogy

Like a car sold as a bare chassis: Thinking Machines shipped the engine and wiring, expecting owners to tune it themselves in Tinker's workshop.

## Why it's emerging now

Thinking Machines Lab, founded by ex-OpenAI CTO Mira Murati, shipped its first open-weights model on July 15, 2026 — a 975B-parameter multimodal MoE trained from scratch and tuned for its own Tinker fine-tuning platform. The launch hit 1,116 Hacker News points in a day, spawning same-day community ports, even as Thinking Machines itself called Inkling not the strongest model available.

## Related terms

- *related:* Thinking Machines Lab
- *related:* Tinker
- *competitor:* DeepSeek V4 Pro
- *competitor:* GLM 5.2
- *competitor:* Kimi K2.6
- *competitor:* Claude Fable 5
- *competitor:* GPT-5.6 Sol
- *competitor:* Nemotron Ultra
- *parent:* Mixture of Experts
- *parent:* Open-Weights Model
- *related:* Multi-Token Prediction
- *related:* MLX

## Sources

1. [Thinking Machines Lab: Inkling — Our Open-Weights Model](https://thinkingmachines.ai/news/introducing-inkling/)
2. [Thinking Machines Lab: Inkling Model Card](https://thinkingmachines.ai/model-card/inkling/)
3. [Hacker News: Inkling launch thread (1,116 points)](https://news.ycombinator.com/item?id=48924912)
4. [Hugging Face: Welcome Inkling by Thinking Machines](https://huggingface.co/blog/thinkingmachines-inkling)
5. [MarkTechPost: Inkling — 975B-Parameter Open-Weights Multimodal MoE](https://www.marktechpost.com/2026/07/15/thinking-machines-lab-releases-inkling-a-975b-parameter-open-weights-multimodal-moe-with-41b-active-parameters-and-controllable-thinking-effort/)
6. [GitHub: PipeNetwork/inkling-mlx — Apple Silicon port](https://github.com/PipeNetwork/inkling-mlx)

---
_Generated by EarlyTerms · https://earlyterms.com/term/inkling_