Thinking Machines Releases Inkling-Small AI Model, Nearing Predecessor's Performance at 1/4 Size

Inkling-Small Introduced

Thinking Machines has launched Inkling-Small, an open-source AI language model, just two weeks after the release of its initial Inkling model. This new model is a 276-billion-parameter multimodal reasoning model, licensed under Apache 2.0. It accepts text, image, and audio inputs, produces text, and supports a context window of up to one million tokens.

Performance and Efficiency

Inkling-Small achieves a score of 40 on the Artificial Analysis Intelligence Index, which is only one point below the original Inkling's score of 41. This is notable because Inkling-Small has 276 billion total parameters and 12 billion active parameters per token, while Inkling has 975 billion total parameters and 41 billion active parameters per token. Artificial Analysis reported that no other open-weight model of Inkling-Small's size or smaller scored higher on the index.

Implications for Enterprises

The primary benefit of Inkling-Small for enterprises is its reduced size and computational demands. Developers can achieve comparable capabilities to the larger Inkling model while significantly lowering compute requirements, inference costs, and deployment footprints. While still too large for personal devices, its smaller scale makes it more practical for organizations with some, but not extensive, GPU resources.

Availability and Pricing

Thinking Machines has made the full weights of Inkling-Small available on Hugging Face and integrated support for fine-tuning through its Tinker model training API. For a limited time, the company is offering a 50% discount on API pricing for the standard 64K-context Inkling-Small model, setting rates at $0.58 per million prefill tokens, $1.44 per million sampled tokens, and $1.73 per million training tokens. Cached prefill requests are priced at $0.116 per million tokens, and a 256K-context variant is also available at different rates.