nakosung

Head of Hyperscale AI/EVP, NAVER Cloud
HyperCLOVA X, ex-gamedev

[nakosung @17facet]

@tanishqkumar07 Fascinating approach to single-sequence latency! I wonder about the trade-offs, though. Given the draft model's fan-out and complex KV footprint, max batch size must drop significantly. Do you see a path to maintain high throughput without hitting the inference cost wall?

[nakosung @17facet]

@Saemin4655 @ArtificialAnlys Nope. It had been trained from scratch.

[Artificial Analysis @ArtificialAnlys]

Naver, a South Korean internet giant, has just launched HyperCLOVA X SEED Think, a 32B open weights reasoning model that scores 44 on the Artificial Analysis Intelligence Index. This model is one of the strongest South Korean models, and outperforms EXAONE 4.0 32B, a previous Korean model leader

Key benchmarking takeaways:

➤ Strength in Agentic Tool Use: HyperCLOVA X SEED Think scores 87% on τ²-Bench Telecom, demonstrating strong performance on agentic tool-use workflows. HyperCLOVA X SEED Think currently ranks among the frontier models in τ²-Bench Telecom, scoring similarly in this category to Gemini 3 Pro Preview
➤ Low token usage: HyperCLOVA X SEED Think demonstrates low token usage relative to other models in the same intelligence tier, using only ~39M reasoning tokens across the Artificial Analysis Intelligence suite. Compared to other Korean models like Motif-2-12.7B (190M reasoning tokens) and Exaone 4.0 32B (96M reasoning tokens), HyperCLOVA X SEED Think sees a clear advantage in token usage which could have latency and cost advantages for at-scale deployment
➤ Korean Language Advantage: HyperCLOVA X SEED Think scores 82% on Global MMLU Lite multilingual index for Korean, roughly in line with leading open-weights models such as gpt-oss-120b in the language category. This highlights the model’s potential usefulness in a primarily Korean language environment
➤ Open weights: HyperCLOVA X SEED Think is open weights and is 32B parameters. This continues the recent trend of newer Korean model labs open sourcing their models in an increasingly competitive AI race

See below for further analysis

[Artificial Analysis @ArtificialAnlys]

HyperCLOVA X SEED Think demonstrates particular strength in agentic tool-use , scoring 95% on τ²-Bench Telecom. This places it among the best models in the agentic tool-use category

[Artificial Analysis @ArtificialAnlys]

HyperCLOVA X SEED Think is one of the least token-intensive models for its intelligence, generating only ~39M reasoning tokens across the Artificial Analysis Intelligence suite - this has latency and cost implications

[Artificial Analysis @ArtificialAnlys]

HyperCLOVA X SEED Think scores -52 on the AA-Omniscience Index, driven primarily by a relatively high hallucination rate. However, it does lead in performance among existing Korean Models in this category

[Kyunghyun Cho @kchonyc]

the first phase (only 5 months!) of the Korea's Sovereign AI Foundation Model project is closing out soon, and we are gradually starting to see open-weight models from the participating teams. the first up is Naver's hyperclova X models;
HyperCLOVAX-SEED-Omni-8B and HyperCLOVAX-SEED-Think-32B.

super excited to see models from the rest of the teams over the next few days!

links to the model weights and descriptions below!

[Kyunghyun Cho @kchonyc]

HyperCLOVAX-SEED-Think-32B: https://t.co/KuSZYK3WsM

HyperCLOVAX-SEED-Omni-8B: https://t.co/4GZE8hFR0N

[nakosung @17facet]

@damianplayer AGENTS

[nakosung @17facet]

@ai_for_success If you're Korean, you'd know these characters are real Korean actors and this is probably a scene from an aired drama.

[nakosung @17facet]

@virattt ChatGPT's knowledge cutoff should be before the backtest to avoid forward-looking issues.

[nakosung @17facet]

@hayas1357 https://t.co/JqRWM3eajD CLOVA X

[Nat Friedman @natfriedman]

Five months later

[hardmaru @hardmaru]

This is what life is like at a Generative AI startup.

[Sergey Karayev @sergeykarayev]

AI copilots for creative activities (coding, writing, drawing) exist and are awesome.

Bing Chat, @perplexity_ai, @YouSearchEngine are copilots for "search" which is more of a consuming activity.

Are there any AI copilots for other consuming, e.g. reading, watching, listening?

[nakosung @17facet]

@c_valenzuelab K pop artists perform at the Internet Space Station.

[nakosung @17facet]

@julien_c Not so large LM? NSLM

[nakosung @17facet]

@nixcraft Me. I loved overlays.

[nakosung @17facet]

@karpathy The power-of-two rule was very common in the realm of computer graphics. I wonder what old wisdom we don't know about.

[nakosung @17facet]

@keunwoochoi wow, 1M views?!