Thinking Machines Lab offers enterprises a US alternative in open-weight AI

2026-07-16 · Source: Computerworld · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Advanced, short

Summary

Thinking Machines Lab, a San Francisco startup, has launched Inkling, its first general-purpose AI model, offering a US-developed alternative in the open-weight market. Inkling utilizes a mixture-of-experts architecture with 975 billion total parameters, 41 billion active during processing, and supports a 1 million token context window. It was pretrained on 45 trillion tokens across text, images, audio, and video, and is designed for coding, tool use, and multimodal tasks. Benchmarks show Inkling scored 77.6% on SWE-Bench Verified, 74.1% on MCP Atlas, 77.1% on BrowseComp, and 79.8% on IFBench. The model is fine-tunable via the Tinker API platform and available through various inference providers and as full weights on Hugging Face, including a quantized NVFP4 checkpoint. Thinking Machines also previewed Inkling-Small, featuring 276 billion total and 12 billion active parameters. Deploying the full model requires significant GPU infrastructure, such as 2 TB of VRAM for the BF16 checkpoint or 600 GB for the NVFP4 version.

Key takeaway

For AI Architects evaluating open-weight models for enterprise deployment, Inkling offers a US-origin option. This can address regulatory or procurement barriers with non-Western models. You must weigh its multimodal capabilities and controllable reasoning against substantial GPU infrastructure costs. The quantized version alone requires at least 600 GB of VRAM. Consider Inkling-Small for reduced costs and latency, and always retest model safety after any fine-tuning.

Key insights

Inkling offers Western enterprises a US-developed, open-weight, multimodal AI model for domain-specific adaptation.

Principles

• Open weights enable self-hosting.
• Domain adaptation beats generic performance.
• Fine-tuning impacts safety filters.

Method

Developers can fine-tune Inkling through the Tinker API platform using 64,000 or 256,000 token context lengths, then test it in the Inkling Playground.

In practice

• Deploy Inkling for knowledge-intensive copilots.
• Use quantized NVFP4 for lower VRAM.
• Retest safety after model fine-tuning.

Topics

• Open-weight AI
• Multimodal Models
• Mixture-of-Experts
• Model Fine-tuning
• GPU Infrastructure
• AI Governance

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

Related on AIssential

• Thinking Machines Unveils New AI Technology — AI innovation focuses on specialized models, security, integration, and strategic compute infrastructure.
• New Model: Inkling by Thinking Machine on Hugging Face — Multimodal models can achieve robust reasoning across image, text, and audio with simplified encoder architectures.
• Thinking Machines Lab’s First Model Gives Nod to Chinese Open Source AI — Thinking Machines Lab's Inkling model demonstrates leveraging established open-source architectures, specifically DeepSeek V3, for new AI development.
• Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling — Adaptable, open-weight AI models, fine-tuned by organizations, can outperform and be more cost-effective than general-purpose proprietary systems.
• Thinking Machines open sources first multimodal language model, Inkling, focused on low cost and 'resistance to censorship' — Inkling offers a natively multimodal, open-source foundation model with programmatic cost-performance control and resistance to censorship.
• 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.

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 Computerworld.