Can Mira Murati’s New Open Model Beat China? Here’s the Catch

According to ZDNet, the Thinking Machines Lab — founded by former OpenAI CTO Mira Murati — has released its latest model, Inkling 2 Small. While Chinese labs have historically dominated open-weight leaderboards, this new western model aims to shake up the landscape with 594 billion parameters and native image input support.

How does Inkling 2 Small perform in daily work?

Third-party tester Artificial Analysis gave the model a score of 31 on its Intelligence Index, placing it on par with Gemini 3.8 Flash. The model stands out for web and game design, alongside major gains in coding capabilities and automation.

However, running this model locally or via API involves distinct trade-offs. Artificial Analysis notes that Inkling 2 Small is notably slow per task and verbose. Furthermore, at $0.30 per one million input tokens, it sits above the $0.24 median for its class.

How to access and run the model

As reported by ZDNet, Thinking Machines is using a staggered release process. Following security and cybersecurity reviews, the company will make the model's weights available next week.

• Watch for the official weight drop from Thinking Machines next week.
• Evaluate your infrastructure to handle a 594-billion parameter model locally.
• Factor in API token pricing at $0.30 per million input tokens for cloud deployments.

Is it worth it for your workflow?

If your team relies heavily on western open-weight models for coding and web design, Inkling 2 Small marks a strong step forward, especially with its hallucination rate dropping sharply from 61% to 27% compared to its predecessor. However, if speed and lean token budgets dictate your Monday morning operations, the model's verbosity and premium pricing require careful budgeting before a full rollout.