Complete Guide to Thinking Machines Inkling

2026-07-20 · Source: Analytics Vidhya · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Advanced, long

Summary

Thinking Machines Lab released Inkling on July 15, 2026, its first general-purpose open-weights foundation model. This multimodal Mixture-of-Experts (MoE) model features 975 billion total parameters, 41 billion active parameters, and a 1 million-token context window. Licensed under Apache 2.0, Inkling is designed for customization and fine-tuning across various domains, including multimodal reasoning, agentic AI, coding, and audio/vision tasks, rather than solely for benchmark leadership. It was trained on 45 trillion tokens using NVIDIA GB300 NVL72 systems and incorporates a novel self-fine-tuning capability. While demonstrating strong performance in mathematical reasoning, visual understanding, and agentic coding, Inkling shows weaker results in factuality and long-horizon agentic coding. Access is available via Tinker Playground, Hugging Face, hosted providers like Together AI, or self-hosting, with a smaller Inkling-Small preview also announced.

Key takeaway

For Machine Learning Engineers evaluating foundation models for enterprise deployment, Inkling offers a highly customizable, open-weights multimodal solution. Its 975B-parameter MoE architecture and self-fine-tuning capability allow deep integration into specific workflows. You should prioritize its flexibility for agentic coding and multimodal tasks. However, for high-stakes factual applications, you must integrate retrieval and verification tools due to its factuality weakness.

Key insights

Inkling is an open-weights multimodal MoE model prioritizing customization and domain-specific fine-tuning over benchmark supremacy.

Principles

• Sparse MoE architectures reduce compute costs.
• Hybrid attention balances efficiency and context.
• Self-fine-tuning automates model customization.

Method

Inkling's self-fine-tuning workflow involves an agent defining behavior, generating synthetic data, writing evaluation logic, launching a Tinker run, and evaluating the updated model.

In practice

• Use Inkling for multimodal agentic coding.
• Pair with retrieval for high-stakes factuality.
• Adjust reasoning effort for optimal performance.

Topics

• Multimodal AI
• Mixture-of-Experts
• Open-weights Models
• Agentic AI
• Self-Fine-Tuning
• Enterprise AI

Best for: AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Mira Murati’s Thinking Machines drops Inkling, an open-weights model anyone can access — Open-weight multimodal models like Inkling offer customizable, cost-effective alternatives to proprietary AI systems.
• Ex-OpenAI CTO Murati's Thinking Machines drops Inkling, a 975B parameter model that leads US labs but trails China — Inkling leads U.S. open-weights models in agentic tasks and efficiency but struggles with factual accuracy and cost.
• Murati's 975B Model Fine-Tuned Itself on Launch Day — and Cracked the Top US Open-Weight Spot — Inkling, a 975B-parameter model, demonstrated autonomous self-fine-tuning to adapt its behavior post-training.
• Thinking Machines Lab drops Inkling & Meta’s Muse Spark 1.1 — The AI landscape is shifting from pure benchmark leadership to models emphasizing customization, efficiency, and interpretability.
• Kimi 3, Inkling: These Two Open Models May Change “Everything” — The release of Kimi K3 and Inkling, featuring massive scale and open weights, signals a potential paradigm shift in AI model utility.
• 🗞️ Mira Murati's Thinking Machines Lab drops massive open-weight AI model, Inkling, Apache 2.0 license, no restrictions. — Agentic AI models and efficient inference techniques are rapidly advancing, driving both performance and cost-effectiveness.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Analytics Vidhya.