Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling

2026-07-15 · Source: AI News & Artificial Intelligence | TechCrunch · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has released Inkling, its first open-weight AI model. This mixture-of-experts system features 975 billion total parameters, utilizing about 41 billion per task, and was trained on 45 trillion tokens across text, image, audio, and video, though outputs are currently text-only. Inkling is designed for calibrated answers, flagging uncertainty, and allows users to adjust "thinking effort" for speed. The company's core strategy is that adaptable, fine-tuned AI will outperform one-size-fits-all models. While not claiming best-in-class performance, Inkling demonstrated 84.7% accuracy on financial reasoning tests for Bridgewater Associates, costing roughly a fourteenth of proprietary models. Thinking Machines developed Inkling in about nine months and plans revenue through its Tinker customization platform and hosting ecosystem, not direct model sales.

Key takeaway

For AI/ML Directors evaluating model deployment strategies, Inkling's release underscores the viability of open-weight, adaptable AI. You should explore fine-tuning open models on your organization's specific data to achieve specialized performance and potentially reduce operational costs, as demonstrated by the Bridgewater Associates project. Be prepared to invest in machine learning talent for effective customization and safety oversight.

Key insights

Adaptable, open-weight AI models, fine-tuned by organizations, can outperform and be more cost-effective than general-purpose proprietary systems.

Principles

• Organizations adapting AI for specific expertise yields superior performance.
• Open-weight models offer cost advantages over proprietary subscription models.
• Calibrated answers, including uncertainty flagging, enhance AI reliability.

Method

Inkling employs a mixture-of-experts architecture, drawing on a fraction of its 975 billion parameters per task, trained on 45 trillion multimodal tokens, and fine-tuned via a customization platform like Tinker.

In practice

• Fine-tune open-source models on proprietary data to achieve specialized performance.
• Consider open-weight models to reduce operational costs compared to API-based solutions.
• Implement "thinking effort" controls in custom AI for speed-accuracy trade-offs.

Topics

• Open-weight Models
• Mixture-of-Experts
• Multimodal Training
• Model Customization
• Enterprise AI
• AI Cost Optimization

Best for: Entrepreneur, CTO, VP of Engineering/Data, Machine Learning Engineer, Director of AI/ML, AI Engineer

Related on AIssential

• The New Enterprise Battle Over Who Owns the Model — Open-weight models like Inkling are driving an enterprise shift towards data sovereignty and custom fine-tuning solutions.
• The New Open-Weights Leader, Big AI’s Political Influence, Predicting Illness, Faster Reasoning — Open-weights AI models are rapidly closing the performance gap with proprietary frontier models, offering high-performance, customizable alternatives.
• Thinking Machines Lab offers enterprises a US alternative in open-weight AI — Inkling offers Western enterprises a US-developed, open-weight, multimodal AI model for domain-specific adaptation.
• Open-Weight AI Models — Open-weight AI models offer direct control, customization, and cost-effectiveness for production workloads.
• Companies turn to Chinese AI models to cut costs — Companies are shifting to Chinese AI models for cost savings and flexibility over premium Western alternatives.
• Open weights are quietly closing up - and that's a problem — The erosion of open weights LLM availability threatens market competition and affordability, shifting power to a few frontier labs.

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 News & Artificial Intelligence | TechCrunch.