Inkling 2 Small
Thinking Machines released Inkling 2 Small, a 594B total and 17B active parameter model trained on 12T tokens in 68 hours, reported to beat Muse Spark.
The Latent Space newsletter title gives the size as 17A-594B (17B active of 594B total), a training run of 68 hours on 12T tokens, and a result ahead of Muse Spark.
- Date
- Thursday 5 November 2026
- Lab
- Thinking Machines
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| Parameters (active / total) | 17B / 594B 594B-total / 17B-active multimodal MoE | company |
| Training tokens | under 12T page says pretrained on <12T tokens; the main pretraining phase was 8.4T tokens in native NVFP4 | company |
| Training time | 68 hours pretraining on GB300s, per Soumith Chintala | company |
| Artificial Analysis Intelligence Index | 31 announced score at launch | third-party |
Sources
- latent.space/p/ainews-inkling-2-small-17a-594b-trained-in-68-hours-on-12t-tokens-beating-m
- thinkingmachines.ai/news/inkling-2-small/
This record was checked and corrected against its sources on 6 November 2026. How we check