Thinking Machines Lab has unveiled Inkling 2 Small, a 594-billion-parameter model that matches top Chinese open-weight AI systems in intelligence and coding ability, signaling a shift in the open AI model competitive landscape.
- Inkling 2 Small holds 594 billion parameters and supports image inputs with text outputs.
- Ranks tied with Gemini 3.8 Flash on Artificial Analysis AI Index with a score of 31.
- Weights set for release next week after security and safety checks.
What happened
Thinking Machines Lab, led by former OpenAI CTO Mira Murati, launched its Inkling 2 Small model, a powerful open-weight AI system with 594 billion parameters. The model supports image inputs but outputs only text, emphasizing applications in web and game design and coding proficiency. Testing reveals substantial improvements over the previous version, highlighted by a 41% gain in automation and a 20% increase in coding capabilities.
Third-party evaluation by Artificial Analysis places Inkling 2 Small on par with leading models like Gemini 3.8 Flash, achieving a composite AI Index score of 31. This positions the model competitively against prominent Chinese models and marks a notable advancement for Western AI startups focusing on open-weight releases. The company plans to share the model weights publicly next week after thorough cybersecurity and safety reviews.
Why it matters
Chinese AI firms have dominated the open-weight model space for the past several years, offering high-performance options that challenge expensive closed models from major US labs like OpenAI and Anthropic. This has caused many investors and enterprise clients to reconsider costly contracts with these providers. Thinking Machines' progress represents a strategic Western response capable of shifting market dynamics by providing competitive open models.
Inkling 2 Small’s release may reduce the gap between Western and Chinese AI innovations in open-weight models, especially notable given its coding strengths and broader intelligence scores. The planned public weight release fosters transparency and wider community engagement, which is critical amid growing concerns over AI safety and security. However, the model's relatively high cost per input token and slower task performance highlight ongoing trade-offs.
What to watch next
Market and technical observers should monitor the upcoming public release of Inkling 2 Small's weights and subsequent adoption by developers and enterprises seeking open AI models. The model's real-world application performance, especially in coding and automated workflows, will be key to assessing its impact relative to competitors from China and Western labs like Mistral and OpenAI.
Additionally, pricing strategies and enhancements addressing the model’s verbosity and execution speed will be important factors influencing uptake. Broader industry trends around open weights releases amid rising AI security and safety scrutiny could also shape the competitive landscape as more labs balance openness with protective measures.