Key takeaways
- Thinking Machines Lab released Inkling 2 Small on November 5, a mixture-of-experts model with 594B total and 17B active parameters.
- It is available only through Tinker for now, with Apache 2.0 weights promised next week after safety testing.
- The benchmark scores in the announcement are the lab's own, and the lab says the model trails closed models such as GPT-6.1 Sol on alignment evaluations.
Thinking Machines Lab released Inkling 2 Small on November 5. It is a mixture-of-experts model, which runs only a slice of its network for each token: 594 billion parameters in total, 17 billion active at a time. For now it is available only through the lab's Tinker platform, with closed weights.
The company says it will publish the full weights under an Apache 2.0 license next week, after external safety testing and red-teaming. Until then, developers can run inference in the Tinker Playground and use Tinker for fine-tuning and sampling. The announcement gives no per-token price.
The model takes image input but not audio, and it supports variable effort levels. Thinking Machines lists its own scores on the announcement page: 72.1% on DeepSWE v1.1, 86.3% on SWE-Bench Pro v2 (Hard), 28.7% on HLE-Diamond and a 31 on the Artificial Analysis Intelligence Index. These are the lab's numbers, not independent results.
The company is open about the weak spots. It says the model is not the most succinct communicator, does not always pick the best tools in codebases with many skills, and is less tuned for chat. On alignment evaluations, it says Inkling 2 Small still trails closed models such as GPT-6.1 Sol.
What to watch
An Apache 2.0 license would let companies run and modify a model of this size without asking the lab. The weights are the part that matters, and they are still a week away and conditional on the safety review. Our explainer on what mixture of experts means covers why a 594B model can run with 17B active parameters.
Sources
- Introducing Inkling 2 SmallThinking Machines Lab
