Thinking Machines' Inkling-Small Beats Its 975B Teacher at a Quarter of the Cost
Thinking Machines releases Inkling-Small, a 12B-active MoE that beats its larger sibling on reasoning at a quarter of the compute cost
- Inkling-Small released today: 276B total / 12B active MoE, open weights on Hugging Face under Apache 2.0.
- Beats its larger sibling on reasoning: 31.6% on HLE vs Inkling's 29.7%; 80.2% on SWEBench-Verified vs Inkling's 77.6%.
- Trained with on-policy distillation: Inkling served as teacher; two additional weeks of agentic coding RL added after distillation.
- Natively multimodal, encoder-free: Processes text, images, and audio jointly; Python-augmented visual reasoning for charts and documents.
- Available now on Tinker Playground: Chat and fine-tuning live today with a limited-time discount on both Inkling models.
- Factuality gap remains: SimpleQA Verified 20.6% vs Inkling's 43.9% -- the larger model still wins on grounded factual recall.
Inkling-Small is out, and it is doing something that does not happen very often: the smaller model is beating the one it was distilled from. Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, just released the full weights of Inkling-Small -- a Mixture-of-Experts (MoE) model with 276B total parameters but only 12B active at inference time, trained to match or exceed its 975B-parameter sibling at roughly a quarter of the compute cost.
A model that learned from its teacher's mistakes
The backstory matters here. Inkling-Small began training after its larger counterpart, which gave the team a chance to improve the training process before the smaller model was ever run. That head start compounded in several ways:
- The team made changes to Inkling-Small's pre-training data mix and machine learning recipe.
- On-policy distillation -- a technique where the student model learns by generating its own outputs and then being corrected by the teacher in real time, rather than just imitating a static dataset -- was used with Inkling as the teacher model.
- Starting from an earlier checkpoint, the team continued scaling agentic coding RL for two more weeks.
With these improvements, Inkling-Small surpassed Inkling on reasoning and agentic coding benchmarks. Inkling maintains an advantage on knowledge coverage and factuality. The factuality gap is real: Inkling scores 43.9% on SimpleQA Verified versus Inkling-Small's 20.6%, so if your use case is grounded factual recall, the larger model still wins.
The numbers that matter
On Humanity's Last Exam (HLE) -- a benchmark of extremely hard, expert-level questions across science, math, and humanities -- Inkling-Small scores 31.6%, ahead of Inkling's 29.7%, and the advantage holds at every thinking budget. That last part is key: it is not just a cherry-picked single data point.