1 hour ago
18
Perplexity Computer has rolled out OpenAI’s GPT-5.6 model family across its platform, making Terra the default engine for subagents and Luna the go-to for automations. The integration, completed by July 12, was just three days after OpenAI made GPT-5.6 generally available on July 9.
What GPT-5.6 actually brings to the table
OpenAI’s GPT-5.6 isn’t a single model. It’s a three-tier family designed for different workloads and budgets.
Sol sits at the top, built for complex reasoning and coding tasks, priced at $5 per million input tokens and $30 per million output tokens. Terra occupies the middle ground, offering competitive performance at $2.50 input and $15 output. Luna is the budget option at $1 input and $6 output, optimized for speed over depth.
Perplexity selected Terra for its subagent workflows, where multiple AI agents collaborate on research, coding, and project management tasks. Luna handles the repetitive automation layer, the kind of work where raw speed matters more than nuanced reasoning.
Why crypto should be paying attention
Projects building decentralized AI agent frameworks are competing for the same fundamental use case that Perplexity just upgraded: orchestrating multiple AI models to complete complex, multi-step tasks. The difference is that Perplexity can integrate GPT-5.6 in three days. Decentralized alternatives face additional latency from consensus mechanisms, token-gated access layers, and smart contract overhead.
The tiered pricing model also creates a benchmark problem for crypto AI projects. When Luna can handle automation tasks at $1 per million input tokens through a centralized API, decentralized compute networks need to articulate why their cost structure, which often includes gas fees and token staking requirements, delivers enough additional value to justify any premium.
The multi-agent workflow race
Rather than relying on a single monolithic model for every task, Perplexity routes different parts of a workflow to different models based on complexity and cost. Research tasks might use Terra’s balanced capabilities. Simple data formatting or scheduling gets routed to Luna. If a task requires deep reasoning, Sol is presumably available as an escalation option.
One area where crypto AI projects may find durable advantage is in the economic layer itself. When Luna charges $1 per million input tokens, that revenue flows to OpenAI. In a decentralized equivalent, it could flow to a distributed network of GPU providers, with token holders governing pricing and allocation.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·