Perplexity is quietly making a big move under the hood: Claude Opus 4.7 is now the default “brains” coordinating its new Computer feature, and it is rolling out to Perplexity Max users across web, iOS, and Android.

In simple terms, Computer is Perplexity’s background worker – you tell it the outcome you want, and it spins up a swarm of AI sub-agents to research, code, write, call APIs, and move projects forward for hours or even months. Until now, that orchestration layer was handled by Claude Opus 4.6, which decided what needed to be done, in what order, and which specialist models to call for each step. Swapping that coordinator for Opus 4.7 means the “project manager” model has just been upgraded to something that is better at reasoning through long, messy workflows and keeping many moving pieces in sync.

Anthropic designed Opus 4.7 specifically for high-stakes, complex work like professional software engineering, multi-step agentic workflows, and long-running enterprise tasks. It can break down big goals into smaller steps, adapt how much “thinking time” it uses based on difficulty, and verify its own outputs before returning a result, which is especially useful when Computer is left alone to run for hours. Early reports highlight stronger coding, better vision for reading dense screenshots and diagrams, and tighter instruction-following with more robust safety guardrails than earlier Opus versions.

This is why the choice of orchestration model matters more than it might seem. Perplexity Computer does not rely on a single large model to do everything; it coordinates a roster of about 19 different models, routing each subtask to whichever one excels at it – Claude for coding and orchestration, Gemini for deep research, GPT-5.x for long-context recall, and lighter models like Grok for quick, low-latency work. The orchestrator is the layer that plans, delegates, and decides when to spawn sub‑agents, when to ask you for input, and when to just keep going, so improvements in its reasoning and reliability can translate into more stable, hands-off automation for users.

On the user side, nothing about the interface really changes – you still describe what you want, start a Computer run, and check in as it works – but under the surface, Opus 4.7 is now the one steering those runs. For Perplexity Max subscribers, the same model is also exposed directly as an option in the normal Perplexity experience on web and mobile, so you can tap into the upgraded reasoning engine even outside of full Computer workflows. The trade-off, as some early commenters have pointed out, is cost: using a premium orchestrator is more expensive for Perplexity, and access is currently gated to the higher-tier Max plan, not the cheaper Pro offerings.

Stepping back, this shift underscores a broader trend in AI tooling: the real product is increasingly the orchestration layer, not any single model. As models specialize – some better at coding, others at research, others at long context – the value moves to systems that can intelligently combine them and keep them working in the background like a reliable digital coworker. By putting Claude Opus 4.7 in charge of that coordination, Perplexity is betting that smarter planning and more dependable long-running agents will matter more to users than raw benchmark scores on any one model.