Thinking Machines introduces Inkling Small: a bet on efficiency, not size

Thinking Machines, the AI lab of former OpenAI CTO Mira Murati, has released its second model, Inkling Small. This is an open-weights reasoning model that, according to Artificial Analysis, scores 40 points on the Intelligence Index—one point less than the larger Inkling model (41), even though it is less than three times smaller in size (276 billion parameters, 12 billion active). No open model of the same or smaller size performs better, according to AA. Inkling Small outperforms Inkling on several coding and reasoning tests, such as Humanity's Last Exam (32% versus 30%) and GPQA Diamond (89% versus 87%). However, it lags behind on agentic tasks and factual knowledge tests, but is more token-efficient: in testing, it requires an average of 24,000 output tokens per task compared to 45,000 for DeepSeek V4 Flash and 78,000 for GPT-5.4 mini. The model handles text, image, and voice inputs, has a context window of 256,000 tokens, and is distributed under the Apache-2.0 license. Weights are available on Hugging Face, and through the Tinker Playground, the model can be used and fine-tuned directly in the browser. Thinking Machines positions its models primarily as a foundation for further fine-tuning on proprietary data, and some see this as the next frontier in AI development.