Genspark's OpenAI Realtime Partnership

This partnership matters because voice is one of the few AI features where being early actually changes the product, not just the marketing. Realtime API let Genspark ship speech to speech agents that can listen, respond, and act inside one live session, instead of stitching together separate speech recognition, language, and text to speech systems. For a 20 person team racing into Japan and broader Asia, that shortens build time, improves latency, and helps Genspark turn voice from a demo feature into a usable workflow layer inside Super Agent.

• OpenAI described the Realtime API as a way to build low latency multimodal speech experiences without chaining multiple models, and said it had tested the product with a small set of partners before public beta. OpenAI later said Genspark was among the first to launch voice experiences on it, which points to earlier feedback loops and faster product iteration.
  1 openai 2 openai
• Genspark already uses a coordinator plus nine models and 80 plus tools to turn prompts into outputs like slides, spreadsheets, calls, and media. Realtime voice fits that architecture well because a spoken request can move directly into tool use, like making a phone call or drafting formatted text through Speakly, instead of stopping at transcription.
  3 sacra
• The clearest comparable is the voice stack market. Infrastructure players like Vapi and Cartesia sell the pipes for building voice agents, while Genspark bundles voice inside a broader no code work product. That means privileged model access is more valuable for Genspark as product differentiation than as a standalone API business, especially in markets where local rivals lack frontier model partnerships.
  4 sacra 5 sacra 6 sacra

The next step is that voice stops being a separate feature and becomes a default interface for agentic work. If OpenAI keeps expanding Realtime capabilities, Genspark can use that head start to make multilingual calling, dictation, and live task execution feel native across regions, while competitors still piece together slower, more brittle voice stacks.