Thinking Machines releases Inkling 2 Small's 594B weights under Apache 2.0
For the head of model risk at a regional bank
Thinking Machines Lab released the weights of its Inkling 2 Small model on Hugging Face on Nov. 12, under the Apache 2.0 license, according to The Specialty News and Snowflake.13
The lab announced the model on Nov. 5 and first ran it only on its Tinker service while outside testers probed it for safety problems, then opened the weights a week later, according to The Specialty News.1 The Hugging Face model card says the model is released with open weights to support research, fine-tuning and integration into third-party products by downstream developers.2
The model card lists 594B total parameters and 17B active, with each token routed to 8 of 384 experts, a context window of up to 1M tokens, and text and image inputs with text output.2
The main NVFP4 checkpoint requires at least 500GB of aggregated VRAM, for example 2x NVIDIA B300 or 4x NVIDIA H200 GPUs, and the MXFP8 checkpoint requires at least 800GB, according to the model card.2
In the evaluations table on its Hugging Face model card, Inkling 2 Small scores 30.6 on the AA Index (v5.0), against 32.6 for DeepSeek-V4.1-Flash-1017 and 27.8 for GLM-5.3-Flash. The same table lists 90.2% on BrowseComp and 14.1% on Sudo L7 for Inkling 2 Small.2
The lab says the model is not the most succinct communicator and does not always pick the best tool in codebases with many skills, The Specialty News reported; the lab's showcase had the model raise a 4B model's forecasting score in 4.5 hours for $426.1
Snowflake said in a blog post dated Nov. 12 that the model is available in Cortex Training (public preview) for post-training and inference. The model card lists API access through providers including NVIDIA, Fireworks, Together, Baseten, Nebius, Modal and Cloudflare.32
Banks can now download and run a large model inside their own infrastructure, which keeps customer data in-house and removes a vendor from the data path. The hardware floor is a multi-GPU cluster, and each fine-tuned version is a new model that needs its own validation under model risk rules.
Our take. Apache 2.0 means you can run Inkling 2 Small without asking anyone. It also means there is no one to ask. Outside red-teamers had a week; your validation file starts empty.
Action. Model risk: Add self-hosted open-weight models and each fine-tuned version to the model inventory, with validation owners assigned before any download.
What our records say
NVIDIA designs GPUs and AI systems and sells them with a licensed software stack for training, customising and serving models, and banks run the stack in their own data centres or rent it in the cloud; bank customers include BNY, American Express, Capital One, PNC and, at group level, Santander and Deutsche Bank.1
From our records
- NVIDIAvendor · banks using it (6) BNY, Amex, Capital One · alternatives SambaNova
Sources
- ReportThinking Machines Releases Inkling 2 Small's 594B Weights Under Apache 2.0 · The Specialty News Thespecialtynews
- Primary sourcethinkingmachines/Inkling-2-Small · Hugging Face Huggingface
- Primary sourceInkling 2 Small from Thinking Machines Lab Is Now Available in Snowflake Cortex Training Snowflake