Thinking Machines has introduced Inkling-Small, an open source AI language model that approaches the performance of its predecessor, Inkling, despite being about one-quarter of its size. The new model has 276 billion parameters and uses 12 billion active parameters per token, compared to Inkling’s 975 billion parameters and 41 billion active parameters.
Inkling-Small’s Performance
Inkling-Small has achieved a score of 40 on the Artificial Analysis Intelligence Index, just one point behind Inkling’s score of 41. The model also surpasses Inkling on several benchmarks, including SWE-bench Verified, Terminal Bench 2.1, and SciCode.
The model’s performance is notable given its smaller size, with Thinking Machines reporting that no open-weight model at Inkling-Small’s size or smaller has scored higher on the Intelligence Index. However, Inkling retains a clear advantage on factual knowledge and some agentic tasks.
Architecture and Deployment
Inkling-Small is a sparse Mixture-of-Experts model, with a 42-layer decoder that routes each token to six specialized experts, along with two shared experts that remain active for every token. The model is natively multimodal, accepting text, image, and audio inputs, and producing text outputs.
Licensing and Adoption
Inkling-Small is released under the Apache 2.0 license, which allows organizations to use, modify, fine-tune, redistribute, and commercialize the model, including inside proprietary products, provided they comply with the license’s notice and attribution requirements.
Related: Mastercard updates fraud rules for AI assistants
This licensing approach gives enterprises far more legal flexibility than many custom “open” AI licenses, which may include revenue thresholds, branding obligations, use restrictions, or separate conditions for large-scale commercial deployment.
For legal, procurement, and platform teams, the Apache 2.0 license can materially simplify adoption, although it does not eliminate the need to review acceptable-use policies, data provenance, regulatory exposure, or downstream safety obligations.
Model Development and Deployment
Thinking Machines has shown that it can quickly turn its first large model release into a repeatable engineering process, with Inkling-Small benefiting from an improved pre-training data mix, changes to the machine-learning recipe, and on-policy distillation using Inkling as a teacher.
Thinking Machines has released the full weights on Hugging Face and added support for fine-tuning through its Tinker model training application programming interface (API). At launch, the company is advertising a limited-time 50% discount, bringing API pricing for the standard 64K-context Inkling-Small model to $0.58 per million prefill (input) tokens, $1.44 per million sampled (output) tokens, and $1.73 per million training tokens, with cached prefill requests priced at $0.116 per million tokens. A 256K-context variant is also available at higher rates.
It signals that Thinking Machines is moving towards a cadence in which open-weight multimodal systems can be produced, improved, and deployed more routinely, which could have significant implications for the development and adoption of AI models in the enterprise, particularly in terms of AI pipeline accuracy.
