Today we’re excited to share our collaboration with Upstage, one of Korea’s leading AI companies, to explore how Cerebras Inference can help bring ultra-fast, production-ready AI experiences to more users around the world. Running on the Cerebras Wafer-Scale Engine, Upstage’s Solar 31B can achieve inference speeds of up to 2,000 tokens per second.
Upstage has built a powerful portfolio of enterprise AI products across large language models, document intelligence, and AI workflow automation. Their Solar model family is designed for enterprise-grade language intelligence, with a focus on speed, groundedness, and practical deployment across business workflows.
“We’re excited to work with Cerebras to bring our Solar models to developers with industry-leading inference speeds. Together, we’re making it easier to build fast, production-ready AI applications at scale,” said Sung Kim, CEO of Upstage.
With Upstage’s Solar models running on Cerebras Inference Cloud, developers can unlock industry-leading inference speeds without changing their existing application development workflows.
Julie Choi, Chief Marketing Officer of Cerebras, said, “South Korea is at the forefront of AI innovation, and we’re thrilled to partner with Upstage to help developers and enterprises across Korea and beyond unlock a new level of inference performance.”
Expanding Fast Inference to South Korea
Upstage builds AI for mission-critical enterprise workflows across South Korea’s leading industries, including finance, insurance, healthcare, and manufacturing. Cerebras provides the inference infrastructure that enables advanced AI applications to run faster, respond in real time, and scale with ease.
Together, we’re bringing ultra-fast inference to one of the world’s most advanced AI markets, empowering developers to build multilingual, real-time AI applications with industry-leading performance.
Building real-time AI experiences
Modern AI applications increasingly retrieve information, reason across context, generate structured outputs, and interact with users in real time.
Cerebras Inference is designed for exactly this kind of workload. With OpenAI-compatible APIs, developers can begin building on Cerebras with minimal code changes while accessing the speed of wafer-scale computing. For enterprises, Cerebras offers production-scale inference with dedicated support, custom model capabilities, and infrastructure designed for high-volume deployments.
Here is an example of Solar 31b taking 246 sources of information and executing a deep research query. A comparable query with Sonnet 4.6 with a similar number of sources takes eight minutes.
By combining Upstage’s model and application expertise with Cerebras’ ultra-fast inference infrastructure, we are working toward a future where powerful AI applications are not only more capable, but also dramatically more responsive.
Why speed matters
For many AI applications, faster inference is not just a performance benchmark. It changes what a product can do.
A search experience can synthesize more sources. A voice agent can feel more natural. A research assistant can reason through more context. A business workflow can complete in seconds instead of minutes. An AI product can become interactive enough for users to rely on every day.
As companies move from AI demos to production systems, latency becomes one of the biggest barriers between a promising prototype and a product people actually use. The faster models can generate high-quality responses, the more room developers have to build richer, more capable applications within the same user experience budget. That shift makes inference speed a critical part of the AI stack.
What comes next
This is only the beginning of our work with Upstage. Our teams are collaborating closely to identify high-impact use cases, evaluate technical opportunities, and explore how fast inference can unlock better AI experiences for developers, enterprises, and end users.
We look forward to continuing our collaboration with Upstage to bring more of their models to the Cerebras platform and deliver the world's fastest inference experiences to developers in Korea and around the globe.