Experience Stronger Korean Fluency In The New Solar Llms

Experience stronger Korean fluency in the new Solar Pro & Mini

Upstage Studio — Build and deploy your agent. Learn more → Upstage Studio — Build and deploy your agent. Learn more →

Solutions

Resources

Company

Pricing

Try now Try demo

Contact us

en

At Upstage, we're dedicated to building AI models that truly understand and communicate in  multiple languages, and Korean is one of the most important languages we consider. With the recent update of Solar Pro and Solar Mini (released Apr 22), we're excited to introduce major improvements in Korean fluency, across a wide range of use cases. A leap forward in Korean fluency Korean fluency is more than just grammar—it's about sounding natural, clear, and aligned with the user's intent. Our internal metric, ko-fluency tracks exactly that—and both Solar Pro and Mini saw significant upgrades in this release. Solar Pro : Increased from 74.4 to 82.0 by +7.6 points, resulting in smoother, more natural conversations. Solar Mini : Increased from 81.2 to 82.1 by +0.9 points, a meaningful gain at this scale.

An example of improvement in Korean fluency. Solar Pro:  Even stronger for Korean use cases

Comparison between Solar Pro, Exaone and HCX. Note that Ko-MT-Bench and Ko-MMLU are not publicly available. Ko-MT-Bench measures an LLM’s Korean conversational ability through multi-turn dialogues covering topics like coding, reasoning, mathematics, knowledge, and role-playing. Ko-MMLU is a Korean multitask benchmark designed to evaluate expert-level knowledge, and also one of the key datasets used in the Open Ko-LLM Leaderboard . Solar Pro continues to be our top-tier model for enterprise-grade reasoning and structured output. This update makes it even more powerful for Korean users, including: Major benchmark improvements, including Ko-MT-Bench  from 71.8 → 81.0 (+9.2) Sharper and cleaner outputs: Better summaries, clearer writing, and a more confident tone.

Available now Updated versions of both Solar Pro and Solar Mini are available on: Upstage Console AWS Marketplace On-premises ( Contact Us for more information)

If you're building Korean AI services, or simply want a smarter, more fluent assistant—this release is for you.

Related posts

Browse all articles

Announcements January 26, 2026

Solar Pro 3: Better reasoning at production scale

Delivering stronger reasoning and more precise instruction-following for production workloads. Optimized to maintain the same efficiency while achieving significant gains in the metrics that matter most for real-world deployments.

Announcements July 10, 2025

Solar Pro 2: Fluent. Reasoning. Frontier.

Pushing the frontier in reasoning, tool use, and multilingual performance. Built to power complex tasks and agent-like workflows across domains.

Announcements May 20, 2025

Solar Pro 2 Preview: Small. Powerful. Now with reasoning.

Solar Pro 2 Preview is here: a compact 31B model that rivals 70B giants, now with hybrid reasoning mode. Still small but smarter, and free to use until July 15 on the Upstage Console.

We build intelligence for the future of work—now it’s your turn.

Start building with our API or talk to our team.

Start with the API Talk to our team

Share

At Upstage, we're dedicated to building AI models that truly understand and communicate in  multiple languages, and Korean is one of the most important languages we consider. With the recent update of Solar Pro and Solar Mini (released Apr 22), we're excited to introduce major improvements in Korean fluency, across a wide range of use cases. A leap forward in Korean fluency Korean fluency is more than just grammar—it's about sounding natural, clear, and aligned with the user's intent. Our internal metric, ko-fluency tracks exactly that—and both Solar Pro and Mini saw significant upgrades in this release. Solar Pro : Increased from 74.4 to 82.0 by +7.6 points, resulting in smoother, more natural conversations. Solar Mini : Increased from 81.2 to 82.1 by +0.9 points, a meaningful gain at this scale.

An example of improvement in Korean fluency. Solar Pro:  Even stronger for Korean use cases

Comparison between Solar Pro, Exaone and HCX. Note that Ko-MT-Bench and Ko-MMLU are not publicly available. Ko-MT-Bench measures an LLM’s Korean conversational ability through multi-turn dialogues covering topics like coding, reasoning, mathematics, knowledge, and role-playing. Ko-MMLU is a Korean multitask benchmark designed to evaluate expert-level knowledge, and also one of the key datasets used in the Open Ko-LLM Leaderboard . Solar Pro continues to be our top-tier model for enterprise-grade reasoning and structured output. This update makes it even more powerful for Korean users, including: Major benchmark improvements, including Ko-MT-Bench  from 71.8 → 81.0 (+9.2) Sharper and cleaner outputs: Better summaries, clearer writing, and a more confident tone.

Available now Updated versions of both Solar Pro and Solar Mini are available on: Upstage Console AWS Marketplace On-premises ( Contact Us for more information)

If you're building Korean AI services, or simply want a smarter, more fluent assistant—this release is for you.

At Upstage, we're dedicated to building AI models that truly understand and communicate in  multiple languages, and Korean is one of the most important languages we consider. With the recent update of Solar Pro and Solar Mini (released Apr 22), we're excited to introduce major improvements in Korean fluency, across a wide range of use cases. A leap forward in Korean fluency Korean fluency is more than just grammar—it's about sounding natural, clear, and aligned with the user's intent. Our internal metric, ko-fluency tracks exactly that—and both Solar Pro and Mini saw significant upgrades in this release. Solar Pro : Increased from 74.4 to 82.0 by +7.6 points, resulting in smoother, more natural conversations. Solar Mini : Increased from 81.2 to 82.1 by +0.9 points, a meaningful gain at this scale.

An example of improvement in Korean fluency. Solar Pro:  Even stronger for Korean use cases

Comparison between Solar Pro, Exaone and HCX. Note that Ko-MT-Bench and Ko-MMLU are not publicly available. Ko-MT-Bench measures an LLM’s Korean conversational ability through multi-turn dialogues covering topics like coding, reasoning, mathematics, knowledge, and role-playing. Ko-MMLU is a Korean multitask benchmark designed to evaluate expert-level knowledge, and...