Mira Murati's Thinking Machines Lab released Inkling, a 975B-parameter open-weights model scoring 41 on the Artificial Analysis Intelligence Index. It leads US open models but trails top Chinese open models on several benchmarks.
Key facts
- 975B total parameters, 41B active (MoE).
- Score of 41 on Artificial Analysis Intelligence Index.
- 63% hallucination rate on factual accuracy.
- Pricing: $1.87 per million input tokens.
- Pre-trained on 45 trillion tokens.
Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, has shipped its first production-ready model. Inkling is a Mixture-of-Experts Transformer with 975 billion total parameters, 41 billion of which are active at any given time. It natively handles text, images, and audio with a 1-million-token context window According to The Decoder. The weights are freely available on Hugging Face.
Fine-Tuning as a Business Model
Thinking Machines is positioning Inkling as a flexible base model for customization, not a top-tier general-purpose model. "Inkling is not the strongest overall model available today," the announcement states. The company expects multimodal support, efficient processing, and fine-tuning options via its Tinker platform to differentiate it. Pre-training used 45 trillion tokens of public and synthetic text, images, audio, and video, including data "that may be subject to intellectual property protection." The company used the Chinese model Kimi K2.5 to generate synthetic data.
Benchmark Realities: US Lead, China Gap
According to Artificial Analysis, Inkling scores 41 on the Artificial Analysis Intelligence Index, three points above the prior US open-weights leader, Nemotron 3 Ultra at 38. On GDPval-AA v2, an agentic benchmark, Inkling reaches an Elo of 1,238. However, the model shows a 63% hallucination rate on factual accuracy tests. Pricing starts at $1.87 per million input tokens — higher than comparable Chinese models like DeepSeek v4 Flash max. While Inkling leads US open models, top Chinese open models still beat it on overall performance, underscoring the persistent gap between US and Chinese open-source AI development.
What to watch
Watch for the next Thinking Machines model release and whether fine-tuning adoption on Tinker drives enterprise revenue. Also track if Chinese labs extend their lead on US open models in the next Artificial Analysis Intelligence Index update.
Source: the-decoder.com
