Grok 4.5 has arrived in Perplexity Computer for both sides of the product: Consumer Pro and Max subscribers can use it as the orchestrator behind their tasks, while Perplexity Enterprise organizations now have access too. The update is notable less because it adds another model option, and more because it shows where the AI race is heading: toward agents that can coordinate real work, not merely answer a prompt.

For everyday users, the change sits inside Perplexity Computer, the company’s agentic product designed to take on longer, multi-step assignments. Instead of treating an AI model as a clever autocomplete engine, Computer uses an “orchestrator” to decide what should happen next: break a request into pieces, choose tools, review intermediate results, and keep moving until the job is done.

That behind-the-scenes role matters. A user may ask for a market brief, a research-heavy spreadsheet, a travel plan with constraints, or an analysis that requires browsing, checking sources, and presenting a usable output. The visible result is a finished task. The less visible, more difficult part is maintaining a sensible plan while dozens of small decisions pile up. That is the work Perplexity is now putting in Grok 4.5’s hands.

Perplexity says it tested Grok 4.5 against five other orchestration configurations on its WANDR evaluation and found that the xAI model scored higher than the rest at roughly half the cost of Claude Opus 4.8. That is the company’s own benchmark claim, rather than an independent finding, so it should be treated as a promising product signal rather than a final verdict on which model is “best.” Still, cost and reliability are exactly the two variables that determine whether an AI agent feels practical outside a demo.

The timing is also important. xAI (now SpaceXAI) positions Grok 4.5 as a frontier model for coding, agentic tasks, and knowledge work, with support for tool use including web search, X search, function calling, and code execution. Its published API price is $2 per million input tokens and $6 per million output tokens, though Perplexity’s internal operating costs and product limits will not necessarily map directly to those figures.

For Consumer Pro and Max users, this means Grok 4.5 is not simply being added as another chatbot personality to select from a model menu. It is being used as an orchestrator in Computer, where planning, recovery from mistakes, tool selection, and sustained task execution matter more than a single polished answer. In plain English: Perplexity is betting that Grok 4.5 can be a better project manager for AI work.

Enterprise availability raises the stakes. Perplexity CEO Aravind Srinivas confirmed that Grok 4.5 is enabled for Perplexity Enterprise organizations, extending the rollout from paid consumer plans to teams that may be using Computer for research, analysis, internal knowledge work, and recurring operational tasks.

That kind of rollout is a pragmatic answer to a common enterprise question: why should a company choose one model vendor when the frontier shifts every few months? Perplexity’s pitch is increasingly about model flexibility. A company can stay in one workspace and let Perplexity decide, or allow users to decide, which underlying system is most appropriate for the job. In that model, the platform’s value is not tied entirely to inventing the strongest base model; it lies in connecting models, information, tools, and workflows in a way that gets reliable work done.

It is also a meaningful distribution win for xAI. Grok has largely been associated with its own consumer experience and the X ecosystem, but embedding the model inside Perplexity Computer puts it in front of users who are already accustomed to asking AI to research the web, synthesize sources, and complete work-oriented tasks. That gives Grok 4.5 a new proving ground where speed, consistency, and execution quality will be tested against alternatives in day-to-day use rather than benchmark charts alone.

There are reasons to stay measured. AI agents can still make poor assumptions, misunderstand instructions, get stuck in loops, or produce confidently wrong work. Enterprises, in particular, will need to assess their data controls, permissions, auditing requirements, and the human review needed before using agent outputs in high-stakes decisions. A stronger orchestrator can make an agent more capable, but it does not remove the need for oversight.

Even so, the move captures a broader shift in the market. The battle is no longer just over which chatbot writes the nicest paragraph or solves the hardest isolated puzzle. It is increasingly about which model can manage a messy, real-world assignment economically enough to be used every day. Perplexity’s decision to bring Grok 4.5 to Computer for paid consumers and enterprise teams suggests that the company sees xAI’s newest model as a serious contender in that more demanding contest.