Thinking Machines introduces Inkling 2 Small and plans weights release

Inkling 2 Small has 594 billion total parameters and 17 billion active parameters.[1] Its architecture is a 48-layer decoder-only transformer, with each token routed to eight of 384 experts.[2] The model accepts text and image inputs and produces text outputs, supports variable reasoning effort and has a context window of up to one million tokens.[1][2] Thinking Machines says it designed the model for developer applications including coding assistants, chatbots and agentic systems.[2] Those specifications give teams several factors to consider when assessing it for their own workflows.

At launch, Inkling 2 Small has closed weights and is initially available through Thinking Machines’ Tinker platform.[1] Developers can try it in Tinker Playground and use it for fine-tuning and sampling on Tinker.[1] Thinking Machines intends to release the full model weights the following week, once external safety testing and red-teaming are completed.[1] The planned release is under the Apache 2.0 license, and the company says checkpoints will be downloadable when the weights are released.[1] The weights release is therefore a future plan, not the model’s launch status.

Artificial Analysis says Inkling 2 Small’s score is 18 points higher than Inkling Small’s.[3] It reports that AutomationBench-AA rose from 5% to 46%, while the hallucination rate fell from 61% for Inkling Small to 27% for Inkling 2 Small.[3] Coding comparisons were not uniform: against Claude Sonnet 5.5 (medium), Inkling 2 Small scored higher on FrontierCode 2.0 but lower on Terminal-Bench 4.0.[3] Artificial Analysis also says the model matches Gemini 3.8 Flash (high) and surpasses Xiaomi’s MiMo-V2.6-Pro on its Intelligence Index.[3] These reported results do not establish performance on every application.

Artificial Analysis lists Inkling 2 Small at $0.30 per million input tokens, $1.00 per million output tokens and $0.03 per million cached input tokens.[3] It calculates the model’s average cost at $1.18 per Intelligence Index task.[4] The same analysis reports that an Intelligence Index task takes 18.5 minutes for Inkling 2 Small, compared with a seven-minute median for comparable models.[4] Artificial Analysis says a weights release would make the model the highest-scoring open-weights model developed outside China in its current rankings.[3] That ranking is conditional on the planned release and refers to those rankings, not a universal measure.