Thinking Machines has launched its inaugural AI model, Inkling, an expansive 975-billion-parameter open source system trained from scratch to handle multimodal inputs including video and audio. This release positions the startup as a notable contender among AI leaders like Anthropic and OpenAI.
- Inkling processes video, audio, and text with 975 billion parameters
- The model improves itself through self-fine-tuning
- Thinking Machines aims to decentralize AI development with open models
What happened
Thinking Machines announced the release of Inkling, its first AI model, featuring 975 billion parameters. Unlike many single-focus models, Inkling is designed to understand and analyze multimodal data such as video, audio, and text. The model was trained entirely from scratch and can perform advanced reasoning and coding tasks.
A unique aspect of Inkling’s development is its capability for self-improvement; the model was used to fine-tune and enhance its own performance. Despite not achieving top scores on all popular benchmarks, Inkling delivers competitive results comparable to the best open-weight models originated from other countries, particularly China.
Why it matters
This release cements Thinking Machines’ presence in the competitive AI landscape, where open-source models are increasingly favored for their cost-efficiency and versatility. Open-weight models like Inkling allow broader access and customization, contrasting with proprietary models that often require costly subscriptions or usage fees.
The company promotes a vision of decentralizing AI to empower more developers to create with their own data rather than relying on a handful of dominant tech firms. The founding team’s pedigree, originating from OpenAI leadership, and a record $12 billion seed valuation highlight the startup’s strategic importance in ongoing AI innovation.
What to watch next
Observers will look to see how Inkling performs in real-world applications and benchmarks compared to competing models like Anthropic’s Claude and offerings from OpenAI. Success in varied use cases, particularly those combining video and audio analysis, could broaden adoption.
Additionally, the evolution of Thinking Machines’ tools for model fine-tuning and voice interaction, alongside Inkling’s expanding ecosystem, will be key indicators of the company’s ability to sustain momentum. The open-source approach could inspire more decentralized AI development, challenging the dominance of closed large language models.