Thinking Machines announces Inkling 2 Small, with Apache 2.0 weights promised next week
A Mixture-of-Experts model with image input and variable effort. It launches on Tinker with closed weights; the lab says full weights follow under Apache 2.0 after outside safety testing.
Why it matters: If the weights ship as promised, it joins the small group of large models under a truly permissive license.
- Parameters
- 594B total
- Active per token
- 17B
- Weights size
- Not released yet
- Context
- Not stated as a spec (training used up to 1M)
- License
- Apache 2.0 planned; closed today
- Trained on
- NVIDIA GB300 NVL72, 68 hours of pretraining
| Benchmark | Score | Type | Measured or published by |
|---|---|---|---|
| AA Intelligence Index | 31 | Cited | Thinking Machines, citing Artificial Analysis |
| DeepSWE v1.1 | 72.1% | Self-reported | Thinking Machines |
| SWEBench Pro v2 (Hard) | 86.3% | Self-reported | Thinking Machines |
| FrontierCode v2.0 | 25.4% | Self-reported | Thinking Machines |
Sources: thinkingmachines.ai