whatmodel.sh / HyperCLOVA X SEED 32B Think
HyperCLOVA X SEED 32B Think
↗ share
+ contribute
HyperCLOVA X SEED 32B Think is an open-weights language model from NAVER Cloud, released December 2025. Its published size is 33.31B parameters and a 128k-token context window. Licence: hyperclovax.
Catalogue identity notes
NAVER Cloud's open-weights vision-language thinking model, repo naver-hyperclovax/HyperCLOVAX-SEED-Think-32B @3c2b633 (created 2025-12-23; licence tag other / hyperclovax; 14 safetensors shards, 66.63 GB). Subject of arXiv 2601.03286v1 ('HyperCLOVA X 32B Think', abbreviated THINK). Comparison-only for now: the card's only documented run route is NAVER's OmniServe Docker stack on 4x A100 80GB (clone, convert_model.py, docker compose), not a single command; no maker GGUF/ollama tag or first-party API page verified. New provider string 'NAVER Cloud' (no existing NAVER entry in data/models.json).
Model information sources
Release date ↗ · Weights licence ↗
Card @3c2b633 says 'Parameters : 32B'; safetensors metadata 33,313,410,304 (vision-language model, includes the vision encoder).
$ how it compares
The LLMs ranked either side of HyperCLOVA X SEED 32B Think, out of 828 with a score.
| # | model | score | where |
|---|---|---|---|
| 219 |  GLM-4.5V | 61±2 | api $0.90/M +local |
| 220 |  Llama 3.1 Nemotron Ultra 253B v1 | 61±2 | local |
| 221 | HyperCLOVA X SEED 32B Think | 61±2 | comparison only |
| 222 |  Devstral Small 2 24B | 61±3 | local |
| 223 |  o1-mini | 61±1 | comparison only |
$ what it's good at
agent tasks▁▁▂▆█▆▅▄▄▃▂▁ #143 of 340 · 81% of its score · 5 tests
Beats o3, Gemini 2.5 Pro, DeepSeek V3.2 Exp here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Large, GPT-4o (2024-05-13) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Terminal-Bench Hard lab | 9.9% ✎ | ▂▅▆▂▂▆▂▂█▅▅▂ beats 13/27 | 23% |
| τ²-bench Telecom lab | 65.1% ✎ | ▁·▅█▆▅▂▃▃▂▃▁ beats 52/89 | 19% |
| τ²-bench Retail lab | 71.6% ✎ | ▁····▂▁▃▆█▃▁ beats 30/44 | 16% |
| τ²-bench Airline lab | 58.0% ✎ | ▁▁·▂▁▃▂▁█▅▂▃ beats 21/42 | 13% |
| K-BrowseComp-Verified (Korean web-browsing agent questions, pass@1) lab | 2.3% ✎ | ▃····█··▃█·▃ beats 0/8 | 12% |
web search▃·▂▇▅█▃▄▁▂▁▁ #92 of 92 * · 12% of its score · 1 test
Trails GPT-4.1 nano, GPT-4o mini (2024-07-18), Mistral Small 3.1 24B (2503) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| K-BrowseComp-Verified (Korean web-browsing agent questions, pass@1) lab | 2.3% ✎ | ▃····█··▃█·▃ beats 0/8 | 12% |
$ where to run it
No verified API or download recorded yet · add one
Weights and documentation: huggingface.co/naver-hyperclovax ↗
$ how the score was worked out · HyperCLOVA X 32B Think (NAVER… 87%, K-BrowseComp: A Web Browsing… 12%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 HyperCLOVA X 32B Think (NAVER Cloud, arXiv:2601.03286v1) · lab-reported | 7 | 87% |
| K-BrowseComp: A Web Browsing Agent Benchmark Grounded in Korean Contexts (arXiv:2606.02404v1) · lab-reported | 1 | 12% |
| prior · pull toward the middle | 2% | |
| weighted average | = 61.0 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json