The U.S. AI startup Genspark is expanding its "AI Workspace" ecosystem through partnerships with OpenAI, Anthropic, and Microsoft, combining frontier models, cloud infrastructure, and enterprise tools to widen how knowledge workers use AI agents.

The strategy behind it is the part worth understanding. While most of the headline money in AI has gone to companies racing to build the most capable model, Genspark made the opposite bet: it builds no model of its own. Its wager is that the durable value sits one level up — in the layer that routes each task to whichever model does it best and turns the result into finished work.

A Bet on Orchestration, Not a Model

At a media tour of its Palo Alto headquarters on June 26, Genspark laid out its partnerships and enterprise strategy. Founded in 2023 and valued at $2.6 billion in a recent funding round, the company — led by CEO Eric Jing, a founding member of Microsoft Bing, and CTO Kay Zhu, who worked on Google search ranking — chose not to build its own model. Instead it orchestrates more than 70 AI models to fit each task, from document creation and research to presentations, real-time calls, and agent automation, without making users pick a model. A coordinator takes a user's goal, breaks it into sub-tasks, and routes each to the model best suited to it, then assembles the pieces into a finished deliverable. The logic is that no single model is best at everything, so the value lies in the routing and execution rather than the raw intelligence.

The approach has scaled quickly. By the company's own account, Genspark has reported $250 million in annual run rate within a year, raised a $485 million Series B at a $2.6 billion valuation — with investors including LG and Temasek — and launched "Genspark Claw," billed as an AI employee that executes multi-step tasks.

Three Partners, Three Different Layers

That orchestration-first design is what makes the three partnerships structural rather than decorative; each supplies a different layer of a stack Genspark assembles rather than owns. The company described the roles at the June 26 event.

OpenAI's startups head, Mark Manara, said working with application companies that have real users — rather than just chasing benchmarks — helps steer research, and the two collaborate on real-time APIs, voice interfaces, image generation, and presentations; Genspark uses OpenAI's real-time voice model to let users instruct agents by talking.

Anthropic's collaboration, according to the company, centers on Genspark's super agent and on internal development productivity. Danny Stein, Anthropic's go-to-market lead for AI-native strategy, called Genspark a representative AI-native company, and CTO Kay Zhu said the relationship goes beyond simple API use to early model testing and research-based collaboration, with Genspark using Anthropic's Claude Code to ship features quickly with a small team.

Microsoft, meanwhile, spans infrastructure, product integration, and market expansion: Genspark runs on Microsoft Azure and uses Teams internally, and Microsoft ASEAN president Mayank Wadhwa described embedding Genspark's capabilities into Microsoft 365, Excel, and PowerPoint while jointly targeting enterprise customers in markets like South Korea and Japan.

The Dependency Beneath the Strategy

Genspark plans to deepen its push into Asian enterprise markets on the strength of those three partnerships. But the orchestration bet carries a tension worth keeping in view: an orchestrator's position depends on continued access to the very model makers it routes to — several of which, including its own partners, build competing agent products. And independent measures of how well agentic-workspace platforms actually perform remain limited; none of the fastest-growing players in the category, Genspark included, has published broad task-success or customer-retention data, and the company's scale figures are its own. The partnerships widen what Genspark can do; whether the orchestration layer is where the lasting value settles is the open question its whole strategy is wagering on.

(Disclosure: Anthropic, one of the partners named in this article, is the maker of the Claude models, including the one used to draft it. The partnership details described here, including Anthropic's role, are drawn from Genspark's June 26 media tour and public reporting rather than any non-public information.)

Frequently Asked Questions

What is Genspark?

Genspark is a Palo Alto-based AI startup, incorporated as MainFunc Inc., that builds an "AI Workspace" for knowledge workers. Rather than building its own large language model, it orchestrates more than 70 AI models from providers including OpenAI, Anthropic, and Google, routing each task to the model best suited to it and assembling finished outputs like presentations, documents, spreadsheets, and web apps. Founded in 2023 by former search-and-AI veterans, it pivoted from AI search to agentic AI in 2025 and has grown rapidly, reaching a $2.6 billion valuation in a 2026 funding round, according to company announcements.

How does Genspark's AI Workspace work?

Genspark uses what it calls a "Mixture-of-Agents" approach. A coordinator model receives a user's high-level goal — such as creating an investor presentation — decomposes it into smaller sub-tasks, and routes each sub-task to whichever of its 70-plus models is most capable of handling that specific type of work, from reasoning to coding to image generation. The system then assembles the results into a finished deliverable, so the user describes the outcome they want rather than choosing a model or stitching tools together. Its "Genspark Claw" feature is positioned as an AI employee that can execute multi-step tasks across software.

Which AI models does Genspark use?

Genspark says it orchestrates more than 70 AI models rather than relying on any single one. These include frontier models from partners such as OpenAI and Anthropic, along with models from Google and specialized models for tasks like code, image generation, and data analysis. The company's stated rationale is that no single model excels at everything, so routing each task to the best-suited model produces better results than depending on one. Its partnerships with OpenAI, Anthropic, and Microsoft cover, respectively, frontier models and real-time interfaces; agent and development collaboration; and cloud infrastructure plus enterprise distribution.

Who founded Genspark?

Genspark was co-founded in 2023 by CEO Eric Jing and CTO Kay Zhu, who had previously worked together for years. Jing was a founding member of Microsoft's Bing search engine and later a vice president at Baidu; Zhu worked on AI-powered search ranking at Google before moving to Baidu. Their backgrounds in large-scale search and AI infrastructure are often cited as the reason Genspark's product is built around model orchestration. COO Wen Sang, an MIT PhD who previously founded an enterprise software company, is also part of the founding leadership.