[Former OpenAI CTO's New Model Inkling-Small Open-Sourced: Less Than 30% of Inkling's Parameters, Only 1 Point Performance Difference] According to monitoring by Beating, former OpenAI CTO Mira Murati's Thinking Machines Lab has released Inkling-Small and made its full weights open-source. This is a multimodal reasoning model that supports text, image, and audio inputs. Like Inkling, the model is licensed under Apache 2.0. Inkling-Small has a total of 276 billion parameters, with 12 billion activated per run. In comparison, Inkling has 975 billion total parameters, with 41 billion activated. On the Artificial Analysis intelligence index, the two models score 40 and 41 points, respectively. Interestingly, the smaller model outperforms in certain tasks. In official tests, it surpassed Inkling in programming and reasoning benchmarks such as SWE-bench Verified and HLE. However, Inkling demonstrates better knowledge coverage and factual accuracy. The output cost for Inkling-Small is $1.20 per million tokens, approximately 70% cheaper than Inkling. The full weights have been uploaded to Hugging Face, but the official quantized weights still amount to 171GB, making it impossible to fully load on standard consumer-grade GPUs. [Original Link]