Inkling 2 Small launches on Tinker, with open weights still to come

Thinking Machines introduced Inkling 2 Small on November 5. Its model card describes text and image inputs with text output, plus API access for inference and fine-tuning and an inference-only playground. [2]

Spicy's takeThe immediate significance is a new hosted model developers can evaluate and customize. Self-hosting is a separate milestone: the launch announcement says weights are initially closed and makes next week's Apache 2.0 release conditional on finishing external safety testing and red-teaming. [1]

The pricing page checked November 7 lists Inkling 2 Small serverless inference at $0.30 per million input tokens, $0.03 for cached input and $1.00 for output. It labels this service beta and directs prospective production users to a waitlist. [3]

Spicy's takeFor a deployment decision, compare access conditions alongside token rates. A beta listing and a production waitlist leave an important practical question—whether your intended workload can actually ship on that service. [3]

Spicy's takeThe model card warns about hallucinations, failures to follow instructions and degradation in long multi-turn exchanges. Those disclosed limits belong in any evaluation plan, especially when a model will answer users rather than merely produce a demo. [2]

Spicy's takeSpicyRanks has not independently tested this model. The useful next step is to evaluate it on your own tasks and failure cases; this launch establishes another option to investigate, not a winner by announcement. [2]