AI Search

Perplexity's Decider v1.3 Scores 70 on Artificial Analysis's Decision Index

Perplexity's open decision models got their first independent report card this weekend. Artificial Analysis scored Decider v1.3 at 70 on its Decision Index v2.1, keeping sixth place, while the smaller Decider Flash v1.3 gained four points to 64 and moved up to tenth.

What was scored

Perplexity released Decider v1.3 and Decider Flash v1.3 as open models under the Apache 2.0 licence, at 27 billion and 4 billion parameters, with a 256,000-token context window and text-plus-image input. The models are decision models, built to classify queries and route them, rather than general chat models, and they are served through Perplexity's own API.

Artificial Analysis scored the pair on Saturday, November 14, 2026, and the results were reported on November 15. Decider v1.3 scored 70 on the Decision Index v2.1 and kept sixth place. Decider Flash v1.3 gained four points to 64 and moved up to tenth, thanks to 12-point jumps on both the AA-Symmetry and When2Call evaluations. The bigger model was fourth on AA-Paperclip at 69 and third on EvalSafe-O*NET-AA at 89.

Cheap, and on the cost frontier

The scores matter because of what the models cost to run. Running a million index tasks costs 68 dollars on Decider v1.3 and 34 dollars on Decider Flash v1.3, which puts both models on the capability-versus-cost Pareto frontier in Artificial Analysis's evaluation. In plain terms, they are among the cheapest ways to get decision-making of this quality, which is exactly the trade-off that matters for a search product routing millions of queries.

That trade-off is the reason Perplexity built the models in the first place. An answer engine has to decide, for every query, what kind of question it is, which tools to call, and which sources to trust, and it has to make that decision fast and cheaply. A decision model that scores well while costing a fraction of a frontier model is infrastructure, not a demo, and Perplexity is now selling that infrastructure through its API as well as using it itself.

Why it matters for SEOs

For SEOs, the relevance is indirect but real. The models that decide how a query is classified also decide which sources get retrieved and cited, and Perplexity's decision layer is now open, scored, and cheap enough to run at scale. The more answer engines that route queries through capable, low-cost decision models, the more the quality of the routing, rather than the size of the model, decides which pages get surfaced.

For link builders, the open release is a reminder that the AI search stack is becoming inspectable. When the routing layer is open source and independently scored, the behaviour of the answer engine is easier to study, and the pages that get cited are the ones that match the query's classified intent most closely. Clear, well-structured pages that answer a specific kind of question are what a decision model routes toward, and that is an optimisation target any site can aim at.

What to watch next

The next Decision Index update will show whether Decider v1.3 holds sixth place or climbs, and whether the Flash model keeps gaining. More important for the search industry is whether other answer engines adopt open decision models of their own, or whether Perplexity's approach stays distinctive. The models are served through Perplexity's API, so developers can test them directly, and the cost figures, 68 and 34 dollars per million tasks, give a concrete benchmark for what routing should cost. If the frontier keeps moving toward cheap, specialised models, the advantage will go to the engines that route well, not the ones that spend most, and the pages that benefit will be the ones whose intent is clearest. The benchmark to watch next is the next Decision Index update, and whether the open models keep climbing while staying cheap.

Sources

Sources: Artificial Analysis Decision Index, via The Daily Fledgling, November 15, 2026; Perplexity, November 13, 2026.

Related reading on this site: Google Launches Gemini Decisions API, Powered by New Gemini Decision 1 Model, Anthropic Opens Private Beta for Self-Hosted Claude Managed Agents, Claude Suffers Major Login Outage as Microsoft's Outlook Outage Drags On, SpaceX's AI Group Lays Off Around 60 From Grok Imagine Team as Imagine Tab Vanishes, Chicago Public Schools May Ditch Google Search Over Unblockable AI Overviews, and AI-Generated Images Are Returning Inside AI Overviews.

More from Linkbaits

Turn this playbook into links

Linkbaits helps you create content that earns citations and backlinks. Start with the toolkit and build your first linkable asset this week.

Start free trial