- Size
- 2.4B parameterscounted from the safetensors header on the Hub
- Licence
- custom, read it, NVIDIA Open Model License AgreementOur reading, check the licence: A custom licence we have not read for this model: other makers use the same name for different texts. Read the model's own licence file before you use it.Licence text
- On the Hub since
- 12 Dec 2025the date the repo was created
- Context length
- Not known
- Fine-tune
- of Qwen/Qwen3-VL-2B-InstructQwen3VLForConditionalGeneration
- On Hugging Face
- nvidia/Cosmos-Reason2-2B491k downloads in the last 30 days, 415 likes
Scores
No independent scores for this model in Epoch AI's data. A score is only shown here when Epoch's name for a model matches this repo with certainty; an uncertain match is left out rather than guessed.
Claimed by the maker Claimed
The model card has no claimed scores in a machine-readable form (the card's model-index), so none are shown. The maker may give scores in the card text: read the card.
Other views, which we link to and do not copy: look this model up on Artificial Analysis and the Arena leaderboard.
Run it yourself
Memory needed = weights + context cache + 0.50 GB for the runtime. Context cache at 8k tokens: 0.88 GB, worked out from the model's config (cagataydev/cosmos-reason2-2b-fp8-hf), at 16-bit precision. Choose a machine above to see a verdict for each format. Method
GGUF for llama.cpp, Ollama, LM Studio and others
apolo13x/Cosmos-Reason2-2B-GGUF (converted by apolo13x), 4.6k downloads. Sizes are the real files on the Hub, all parts added up. For image input you also need the projector file mmproj-Cosmos-Reason2-2B-F16.gguf (0.76 GB), not counted below.
| Format | Size | Memory needed | Get it |
|---|---|---|---|
| Q4_K_M | 1.2 GBfile size | 2.6 GB1.2 + 0.88 + 0.50 | Commandsllama.cpp llama-server -hf apolo13x/Cosmos-Reason2-2B-GGUF:Q4_K_MOllama ollama run hf.co/apolo13x/Cosmos-Reason2-2B-GGUF:Q4_K_MDownload hf download apolo13x/Cosmos-Reason2-2B-GGUF --include "Cosmos-Reason2-2B-Q4_K_M.gguf" --local-dir Cosmos-Reason2-2B-GGUF |
| Q8_0 | 2.0 GBfile size | 3.4 GB2.0 + 0.88 + 0.50 | Commandsllama.cpp llama-server -hf apolo13x/Cosmos-Reason2-2B-GGUF:Q8_0Ollama ollama run hf.co/apolo13x/Cosmos-Reason2-2B-GGUF:Q8_0Download hf download apolo13x/Cosmos-Reason2-2B-GGUF --include "Cosmos-Reason2-2B-Q8_0.gguf" --local-dir Cosmos-Reason2-2B-GGUF |
| F16 | 3.8 GBfile size | 5.2 GB3.8 + 0.88 + 0.50 | Commandsllama.cpp llama-server -hf apolo13x/Cosmos-Reason2-2B-GGUF:F16Ollama ollama run hf.co/apolo13x/Cosmos-Reason2-2B-GGUF:F16Download hf download apolo13x/Cosmos-Reason2-2B-GGUF --include "Cosmos-Reason2-2B-F16.gguf" --local-dir Cosmos-Reason2-2B-GGUF |
MLX for Apple silicon
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| MLX 4bitmasato25/Cosmos-Reason2-2B-Arbor-4bit-g32 | 1.8 GBfile size | 3.1 GB1.8 + 0.88 + 0.50 | Download page |
| MLX 8bithzang/Cosmos-Reason2-2B-8bit | 2.5 GBfile size | 3.8 GB2.5 + 0.88 + 0.50 | Download page |
Original weights size from the safetensors header, for vLLM, SGLang, Transformers
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| BF16original weights | 4.5 GBfile size | 5.9 GB4.5 + 0.88 + 0.50 |
hf download nvidia/Cosmos-Reason2-2B --local-dir Cosmos-Reason2-2BOther conversions (7; sizes not read)
- cagataydev/cosmos-reason2-2b-fp8-hf FP8, 65 downloads
- embedl/Cosmos-Reason2-2B-W4A16-Edge2 Other, w4a16, 54 downloads
- vrfai/Cosmos-Reason2-2B-NVFP4 NVFP4, 41 downloads
- cagataydev/cosmos-reason2-2b-fp8-onnx FP8, 22 downloads
- embedl/Cosmos-Reason2-2B-W4A16 Other, w4a16, 13 downloads
- embedl/Cosmos-Reason2-2B-W4A16-Edge2-FlashHead Other, w4a16, 10 downloads
- embedl/Cosmos-Reason2-2B-NVFP4A16 NVFP4, 0 downloads
The Hub tags 14 quantised repos as conversions of this model; we read the most downloaded and show the maker's own first.
Run it by API
No provider on Hugging Face Inference Providers serves this model (as of 19 Nov 2026, 13:58 UTC). Other hosts may; we only list what the router reports.
Not built yet: one key for every open model, with BlockRun, and our own tested-quality score per provider. Until then, precision and measured speed are the best guide to a provider here.
Planned for this page Not built yet
- Try it in private chat
- A link to Solana AI's confidential chat for the models it serves. This model is not listed as served there.
- Fine-tune it
- Fine-tuning a few base models from this page and publishing the result here.
- Creators and earnings
- Pages for people who publish fine-tunes and adapters of this model, and what they earn.
- Fund and bounties
- Bounties for new open models, paid from a public fund. How the fund will work.
- Contribute compute
- How to give compute to open model work, with Akash.
Model data from the Hugging Face Hub 19 Nov 2026, 13:43 UTC. The repo. Found a mistake on this page? Tell us.