What Is Falcon 2? The 11B Open-Weight Series Explained

What is actually in the series

There are two public checkpoints. Falcon 2 11B is a causal decoder-only language model. Falcon 2 11B VLM pairs that same 11B decoder with a CLIP ViT-L/14 vision encoder so it can answer questions about an image as well as text.

Both are released as raw weights on Hugging Face. TII does not run a public Falcon 2 API, so there is no official endpoint to call: if you want Falcon 2 answering prompts, you host it.

The headline numbers

11 billion parameters in both checkpoints.
Trained on more than 5,000 billion tokens of RefinedWeb plus curated corpora.
8,192 token context window in the text config.
Roughly 22 GB of weights in bfloat16, or 7-8 GB at 4-bit quantisation.
Licensed under the TII Falcon License 2.0, an Apache 2.0-based licence with an acceptable use policy.

What Falcon 2 is not

Falcon 2 is not a hosted service, and it is not one of the newer Falcon 3 or Falcon-H1 families. It is also not a chat product out of the box: the text checkpoint is a raw pretrained model, and TII recommends fine-tuning it for most use cases.

Frequently asked questions

Is Falcon 2 free to use?+

The weights are free to download and the licence permits commercial use, but hosting them is not free. You pay for the GPU or the inference provider that runs them.

Can I use Falcon 2 without a GPU?+

Only slowly. A 4-bit quantisation of the 11B weights is about 7-8 GB, so a CPU with enough memory can run it, at speeds measured in a few tokens per second rather than tens.

Sources

falcon2.lol is an independent third-party site. It is not affiliated with, endorsed by, sponsored by or operated by the Technology Innovation Institute, and it does not speak for TII. Every number on this page is attributed to the source listed above.