Thinking Machines unveils 975B-parameter Inkling model in US-China AI race
A new open-weights AI model from Thinking Machines Lab—led by former OpenAI CTO Mira Murati—has entered the global race for frontier intelligence. Named Inkling, the model packs 975 billion parameters and tops U.S. open-weights models on the Artificial Analysis Intelligence Index, yet trails top Chinese open models on certain benchmarks.
Unlike proprietary giants that guard their weights, Inkling is released under open-weights terms, inviting developers to fine-tune it for domain-specific tasks. Pricing is set at $1.87 per million input tokens, positioning it as a cost-effective base for experimentation rather than a plug-and-play replacement for the most powerful closed systems. The model’s multimodal design suggests it can process both text and visual inputs, a growing requirement for modern AI pipelines.
The open-weights gamble
Open-weight releases have become a strategic lever for Western labs seeking to democratize access while competing with China’s tightly controlled ecosystem. Inkling’s scale—nearly a trillion parameters—reflects the industry’s push toward larger, more capable models that can generalize across tasks without fine-tuning. Yet its performance gap on some benchmarks underscores a persistent challenge: China’s open models often lead in specialized evaluations, particularly in reasoning and multilingual settings.
From lab to market
Thinking Machines is positioning Inkling as a foundation for fine-tuning rather than a stand-alone powerhouse. This approach lowers the barrier for startups and researchers who lack the compute to train models from scratch but need high-quality weights to build on. The pricing model, disclosed upfront, signals a shift toward transparent cost structures in a market where opaque pricing and usage limits are common.
Why it matters
The release highlights how open-weights models are becoming a battleground for AI leadership. While scale and transparency matter, the real stakes lie in who can turn open releases into viable ecosystems—attracting developers, driving innovation, and ultimately shaping which models become industry standards. The U.S. may lead on sheer parameter counts, but China’s benchmarks show that performance still trumps scale in critical areas. For developers, Inkling offers a practical entry point, but the broader question remains: can open models bridge the performance gap without sacrificing accessibility?
Source: The Decoder. AI-assisted editorial synthesis — TechnoExpress.