Opera adds deep research agent ODRA to its Neon AI browser

Opera is pushing further into the AI workspace game.

Just three weeks after launching its subscription-based AI browser, Opera Neon, the Norwegian company is rolling out ODRA -- a new Deep Research Agent designed to tackle complex analysis without leaving your browsing session.

ODRA joins Neon’s existing trio of agents (Chat, Make, and Do) as the platform’s fourth dedicated AI helper.

Parallel models and isolated tasks

Instead of just summarizing a single webpage, ODRA acts more like an automated research assistant.

When given a complicated prompt, it breaks the query into smaller pieces, routes them through multiple AI models -- including systems from OpenAI and Google -- and compiles the findings into a structured summary.

Opera uses a server-side architecture to run several research threads in parallel. In internal evaluations using the DeepResearch Bench test, Opera claims ODRA scored near the top among comparable research agents.

All of this lives inside Neon's Task framework, which keeps contexts isolated so agents can work across multiple open pages and chats without mixing up project details.

The Symphony endgame

Managing four different AI agents sounds tedious, but Opera claims that is not the long-term plan.

Opera EVP Krystian Kolondra revealed an internal orchestration system dubbed Symphony. The goal is a master AI layer that directs specialized sub-agents behind the scenes.

As Kolondra put it, if a user eventually has 50 specialized agents available, they shouldn't have to micromanage them. You simply state your intent, and the main system orchestrates the rest.

Opera Neon is currently available by invite only under a subscription model, with broader access planned soon.

Why it matters

Browsers are quietly evolving from dumb web viewers into agentic operating systems.

For power users doing heavy desk research, market analysis, or multi-source synthesis, having deep research tied directly to browser workspaces is genuinely appealing. The context-isolated Task model solves one of the biggest headaches with web-based LLMs: context pollution across different projects.

That said, it is a paid, invite-only tool. If your daily work only requires occasional lookups, tools like Perplexity or ChatGPT deep search are already readily accessible without an additional browser subscription.

We think Opera's real bet isn't just ODRA, but the Symphony orchestration concept. Individual features get cloned fast by Chrome or Edge, but mastering the handoff between dozens of invisible agents could actually make agentic browsing practical. The real test will be whether ODRA's multi-model synthesis is accurate enough to save time, or if users will still spend 20 minutes double-checking its hallucinations.