Mira Murati's Thinking Machines Drops Inkling, a 975B Open Multimodal Reasoning Model
Thinking Machines releases Inkling, a 975B open-weights MoE model with native audio, vision, and controllable reasoning effort
- Thinking Machines Lab releases Inkling, its first open-weights model: 975B total / 41B active MoE, trained from scratch on 45T tokens.
- Natively multimodal across text, image, and audio with an encoder-free architecture; ranks among the strongest open-weights models on VoiceBench (91.4%) and AudioMC (56.6%).
- Controllable thinking effort lets you trade token spend for performance; matches Nemotron 3 Ultra on Terminal Bench at ~1/3 the tokens.
- Full weights available on Hugging Face (Apache 2.0), with support for SGLang, vLLM, llama.cpp, and Transformers.
- Fine-tuning available now on Tinker at 50% discount; API access via Together, Fireworks, Modal, Databricks, and Baseten.
- Inkling-Small (276B total / 12B active) previewed alongside it, with full weights coming soon; matches Inkling on most reasoning benchmarks.
Thinking Machines Lab just released Inkling, its first model trained from scratch with full weights publicly available. Inkling reasons natively over text, images, and audio, balances cost against performance with variable and efficient thinking effort, and is trained to exhibit safe behavior across modalities. This is the first open-weights model from the lab founded by former OpenAI CTO Mira Murati and co-founder John Schulman, and it arrives with a full ecosystem of inference and fine-tuning support baked in from day one.
A big model, built broad
Inkling is a mixture-of-experts transformer with 975B total parameters and 41B active parameters. It supports a context window of up to 1M tokens and was pretrained on 45 trillion tokens of text, images, audio, and video. The MoE design means only a fraction of the model activates per token, keeping inference costs manageable despite the massive total parameter count.
Inkling is not the most performant model available today, closed or open. The team trained it for solid capabilities across the board rather than state-of-the-art performance in a single area, to serve as a foundation for the models they will train in the future. That framing matters: this is a platform play, not a benchmark chase.
The audio story is the real headline
Most multimodal models treat audio as an afterthought, bolting on a speech-to-text layer before the LLM ever sees it. Inkling takes a different approach. The multimodal components are trained from scratch on general-domain data, using an encoder-free architecture for audio and vision inputs. Audio signals are input as discrete dMel spectrograms, while images are encoded as patches of 40x40 pixels using a four-layer hMLP. Both are transformed via a lightweight embedding layer and processed jointly with text tokens.
The results on audio benchmarks are notable for an open-weights model:
- VoiceBench: 91.4% (vs. 94.3% for Gemini 3.1 Pro and 88.8% for Qwen3-Omni)
- MMAU: 77.2% (vs. 82.5% for Gemini 3.1 Pro and 77.5% for Qwen3-Omni)
- AudioMC: 56.6% (vs. 66.8% for Gemini 3.1 Pro, and 24.3% for Qwen3-Omni)
On AudioMC in particular, Inkling more than doubles Qwen3-Omni's score, placing it in a different tier among open-weights models for audio understanding.