AI startup Thinking Machines launches an open-weight AI model

Reuters | July 15, 2026 at 06:13 PM UTC
Bullish 80% Confidence Unanimous Agreement
Read Original Article

Key Points

  • Inkling has 975 billion parameters, making it one of the largest open-weight models available for users to download, run, and customize
  • The model performs competitively against closed models from Anthropic, Google, and OpenAI, particularly excelling at agent-related tasks
  • The release addresses a gap in Western open-source offerings, as businesses have increasingly turned to Chinese models like Alibaba's Qwen as alternatives to expensive proprietary systems

AI Summary

Summary

AI startup Thinking Machines, founded by former OpenAI CTO Mira Murati, launched its first general-purpose AI model called "Inkling" on July 15, 2026. The San Francisco-based company, established in 2025, positions the model as a Western alternative to dominant Chinese open-source AI offerings.

Key Details:

  • Model Specifications: Inkling features 975 billion parameters, making it one of the largest open-weight models available
  • Accessibility: Open-weight structure allows users to download, run, and customize the underlying systems
  • Availability: Accessible through Thinking Machines' platform Tinker (launched October 2025) and other developer platforms

Market Context:

The Western open-source AI ecosystem has lagged behind China, particularly after Meta shifted away from open-source following its disappointing Llama 4 release. This void has driven businesses to adopt Chinese models like Alibaba's Qwen as cost-effective alternatives to expensive proprietary systems.

Performance:

Benchmark comparisons show Inkling remains competitive against closed models from Anthropic, Google, and OpenAI, plus leading Chinese open-source offerings. While established models maintain overall performance advantages, Inkling demonstrates particular strength in agent-related tasks.

Real-World Adoption:

Hedge fund Bridgewater Associates previously used Tinker to build a custom version of Alibaba's Qwen model, achieving better performance than top proprietary models at reduced costs.

Market Implications:

Inkling's launch could reduce Western dependence on Chinese AI models and provide businesses with a domestically-developed open-source alternative, potentially impacting the competitive landscape for both open and closed AI systems while offering cost advantages over proprietary solutions.

Model Analysis Breakdown

Model Sentiment Confidence
GPT-5-mini Bullish 80%
Claude 4.5 Haiku Bullish 75%
Gemini 2.5 Flash Bullish 85%
Consensus Bullish 80%