Mira Murati's Inkling AI Model Tested Shows Mixed Performance

Inkling is a 975‑billion‑parameter open‑source model released by Thinking Machines Lab, founded by former OpenAI CTO Mira Murati. Independent testing by Decrypt found the model excelled in benchmark scores, achieving 74.1% on the MCP Atlas agent‑tool metric and 77.6% on the SWE‑Bench GitHub debugging test, outperforming rivals such as Nvidia’s Nemotron. In everyday tasks, however, Inkling struggled: complex coding prompts produced blank or low‑quality outputs, creative‑writing generated factual errors, and travel‑planning information contained inaccuracies. The model cannot run on typical consumer hardware and is priced at $1 per million input tokens via OpenRouter. It emphasizes privacy and compliance, appealing to organizations that avoid routing data to Chinese servers. The release marks a Western entry into an open‑source AI field previously dominated by Chinese models, but the current version is more suited for enterprise‑level compliance workflows than for routine developer use.