Sendbird Upstage Deal Brings Solar LLM to AI Agents

Sendbird and Upstage have signed a memorandum of understanding to bring Upstage's homegrown Solar large language model into Sendbird's enterprise AI agent platform, Delight.ai. According to Aju Press, which reported the announcement on October 19, the two companies will integrate Solar with Delight.ai to strengthen AI agents used in corporate settings, and will explore joint business opportunities in the sovereign-AI market.

Under the Sendbird Upstage agreement, the partners will first identify scenarios where Solar can be applied inside Sendbird's corporate client operations and validate them in stages, as reported by Aju Press. The evaluation will measure response speed, task accuracy, and customer satisfaction in real service environments, with the findings fed back into improvements on both platforms and into the model itself.

What each side brings to the table

Sendbird will contribute its experience in AI agent design, system integration, testing, and operations for global clients, while Upstage will handle model optimization tailored to corporate environments using Solar's underlying technology. Aju Press reports that evaluation results and related data will be handled in accordance with client contracts and data protection principles.

Biggo Finance adds that testing has been under way since August, and that the partners plan staged pilot tests with Sendbird's enterprise customers before any wider rollout. Sendbird's communications infrastructure is a substantial proving ground: according to the company's own materials, its platform powers more than 7 billion messages a month across 4,000 web and mobile apps, including deployments at companies such as Match, Yahoo, and Rakuten.

The sovereign-AI bet

The partnership lands as demand grows for capable AI models built outside the largest US labs. South Korea has been one of the most active governments backing domestic AI models, treating them as strategic infrastructure rather than just software. With dependence on foreign models from OpenAI and Anthropic running high across the global market, pairing a domestic Korean model with an established agent operations platform targets companies weighing sovereign-AI options, according to Biggo Finance's analysis.

Upstage has been positioning Solar as a national-model contender. The company is part of South Korea's government-backed Independent AI Foundation Model Project, known as Dopamo. Earlier this month it also joined a consortium with SK Telecom, six cybersecurity companies, and three universities to train AI models on real-world security data, as reported by Aju Press and Telecompaper.

The commercialization question remains open. Analysts cited by Biggo Finance note that the key tests are whether pilot results convert into real contract growth, and whether the partners can match foreign models on price and performance. An MOU signals intent; enterprise purchase orders will decide whether the sovereign-AI pitch has teeth.

Why agents care which model runs them

For enterprise AI agents, the underlying model is a cost and control decision, not just a capability one. Agents that run long workflows can make thousands of model calls a day, so token pricing and latency directly shape the economics of deployment. A domestic model option gives buyers negotiating leverage and, in regulated industries, a data-residency story that foreign models cannot easily offer.

Sendbird itself has been pushing agents deeper into customer experience. Its Delight.ai platform recently added Agent Steward, a single AI owner for complex customer issues, alongside Trust OS 2.0 governance and proactive Voice 2.0 outreach supporting more than 100 languages, launched at the company's Delight Spark event, according to a company announcement carried by PR distributors.

That trust theme is running through the whole agent stack. World recently debuted World ID for AI agents to help banks distinguish verified agents acting for real people from anonymous automated traffic, while startups like Opsin are raising venture money to police what enterprise agents do once deployed. Model provenance is becoming part of the same conversation.

If the Sendbird Upstage pilots validate Solar's speed and accuracy inside Delight.ai, Sendbird's corporate clients get a credible alternative to US models, and Upstage gets the enterprise reference customers every model maker needs. Watch the pilot results, not the press release.