Mira Murati’s Thinking Machines drops Inkling, an open-weights model anyone can access

2026-07-15 · Source: AI – SiliconANGLE · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

Mira Murati's Thinking Machines Lab Inc. launched Inkling, its first foundation model, making its full open weights available to developers. This mixture-of-experts model features 975 billion parameters, utilizing about 41 billion per prompt for faster, lower-cost processing. Trained on 45 trillion tokens across text, image, audio, and video, Inkling offers text-only outputs including code and structured data. The model aims to provide a Western open-source alternative to Chinese AI systems, filling a gap exacerbated by Meta's shift to proprietary models. Developed in less than nine months, Inkling was trained on Nvidia's GB300 NVL72 system. It emphasizes accessibility, customization, and multimodal collaboration, allowing developers to fine-tune it via the Tinker API, which launched in October. Inkling achieved 84.7% on financial reasoning benchmarks at less than 10% cost compared to proprietary alternatives.

Key takeaway

For AI Engineers evaluating foundation models, Inkling presents a compelling Western open-weight alternative to proprietary or Chinese systems. You should consider its customizability via Tinker and its multimodal reasoning capabilities to achieve specialized performance at significantly reduced costs. This approach allows you to control infrastructure spend and integrate the model deeply into your software substrate, avoiding compounding switching costs.

Key insights

Open-weight multimodal models like Inkling offer customizable, cost-effective alternatives to proprietary AI systems.

Principles

• Open weights enable deep customization.
• MoE architectures balance speed and cost.
• Base model choice impacts long-term costs.

Method

Thinking Machines developed Inkling as a mixture-of-experts model, trained on 45 trillion multimodal tokens, and offers fine-tuning via its Tinker API, allowing developers to adjust "thinking effort" for speed-accuracy tradeoffs.

In practice

• Fine-tune Inkling on Tinker for tasks.
• Adjust "thinking effort" for speed-accuracy.

Topics

• Inkling
• Open-weight Models
• Mixture-of-Experts
• Multimodal AI
• Tinker API
• Foundation Models

Best for: Machine Learning Engineer, NLP Engineer, CTO, AI Engineer, Director of AI/ML, Tech Journalist

Related on AIssential

• Complete Guide to Thinking Machines Inkling — Inkling is an open-weights multimodal MoE model prioritizing customization and domain-specific fine-tuning over benchmark supremacy.
• 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.
• 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.
• 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.
• 🗞️ 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.
• The Blank Slate AI Strategy — A "blank slate" strategy commercializes open-source AI by offering a free base model and charging for customization services.

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 AI – SiliconANGLE.