Inkling 2 Small Launches on Tinker With Weights Next Week
Thinking Machines Lab put Inkling 2 Small on Tinker on Nov. 5: a 594B MoE with 17B active and 1M-token context. Will Apache 2.0 weights land next week?
Hakkı ALKANPublished 2 min read
Thinking Machines Lab launched Inkling 2 Small on Nov. 5, 2026, opening the model on its Tinker platform for inference and fine-tuning while keeping weights closed for now. In the company blog post, the lab said full weights are planned next week under an Apache 2.0 license after external safety testing and red-teaming.
What Is Inkling 2 Small?
Inkling 2 Small is a sparse Mixture-of-Experts (MoE) multimodal model, meaning each token is routed to a subset of experts instead of activating every parameter. It has 594B total parameters and 17B active, accepts text and images, supports up to 1M tokens of context, and does not support audio. Mira Murati's lab positions it as the first model in its second Inkling generation, aimed at agentic coding, vision, and long-horizon work.
How Much Does Inkling 2 Small Cost on Tinker?
According to the lab's Tinker pricing page, the 64K tier lists $0.30 per million prefill tokens, $0.03 cached, $1.00 sample, and $1.50 train. The 256K tier keeps the same prefill and sample rates while training rises to $3.00. Developers can try the model in Tinker Playground today; downloadable checkpoints are tied to the forthcoming weight release.
How Does Inkling 2 Small Score Against Rivals?
Thinking Machines reports an Artificial Analysis Intelligence Index score of about 31, with 86.3% on SWE-Bench Pro v2 Hard and a DesignArena Website Elo near 1,340 pending the open-weight drop. Training used native NVFP4 on NVIDIA GB300 racks, then a joint reinforcement-learning run of more than 17 million rollouts over nine days. That puts the launch in the same news cycle as other frontier agent tools covered in our GPT-6 Cyber report.
Our Take
The practical story is customization first: teams can fine-tune on Tinker before self-hosting under Apache 2.0. Limits remain: no audio, weaker chat polish, and alignment still trailing top closed models.
For another consumer-facing AI product move this week, see Claude Personal Finance.
Data-center politics around the same AI buildout are covered in OpenAI's Community Compact.
| Item | Detail |
|---|---|
| Model | Inkling 2 Small (MoE) |
| Parameters | 594B total / 17B active |
| Context | Up to 1M tokens |
| Access now | Tinker inference + fine-tune |
| Weights plan | Apache 2.0 next week |
| 64K sample price | $1.00 / 1M tokens |