AI Models on Perspective AI: GPT, Claude, Gemini, Grok, DeepSeek

TL;DR: Perspective AI offers 60+ AI models in one $14.99/mo subscription: text and reasoning flagships from OpenAI (GPT), Anthropic (Claude), Google (Gemini), xAI (Grok), plus GLM, Qwen, Mistral, Kimi, and DeepSeek, alongside image, video, and search-native models. Every model has a card showing strengths, context window, and running cost. Two routing modes: Custom (you pick) and Auto (the platform picks per request). Perspective AI routes to frontier models; it does not host them.

Perspective AI offers 60+ AI models in one $14.99/mo subscription: text and reasoning flagships from OpenAI (GPT), Anthropic (Claude), Google (Gemini), and xAI (Grok), open-weight families like GLM, Qwen, Mistral, Kimi, and DeepSeek, plus image, video, and search-native models. You can chat with any of them now without a separate subscription per lab. This page is the catalog reference: what is in it, how models are labelled and priced, and how the platform decides which model answers you. It sits in our model access section, and the subscription math behind it lives in the AI pricing guide.

One framing matters before the list. Perspective AI is model-agnostic by design: it routes to frontier models, it does not host them. Rather than locking you into one lab, the platform curates the strongest proprietary and open-weight models side by side and lets you move between them freely inside a single interface. What "routes, not hosts" means in practice: when you send a prompt to a GPT or Claude flagship, the request goes to the lab that runs that model and the answer comes back into your thread. The aggregation is honest, and that honesty is the point. You get the real frontier models, not lookalikes with borrowed names, and you get them without maintaining a separate login, bill, and history silo per lab.

The catalog at a glance

The catalog spans the major labs and every modality. Four groups cover it:

Every modality bills into the same credit balance, so a research question, an illustration, and a video clip draw from one pool rather than three separate tools. If you are deciding which text model fits which job, our guide to which AI model you should use breaks that down task by task.

Foundational vs open source: how models are labelled

Each model in the catalog carries one of two labels:

• Foundational: proprietary frontier models such as the GPT, Claude, Gemini, and Grok flagships. These run with their providers; Perspective AI routes your request and returns the answer.
• Open source: open-weight models the network can ultimately serve itself, including the GLM, Qwen, Mistral, Kimi, and DeepSeek families.

The distinction is practical, not cosmetic. It tells you where the model runs, and it explains why you no longer need a lab-by-lab subscription stack. If you currently pay Anthropic just for Claude, or Google just for Gemini, the catalog covers both: see how to use Claude without an Anthropic subscription and how to use Gemini without Google One.

Model cards: what each listing tells you

Every model has a card describing four things:

1. What it is best at, so you can match the model to the task instead of guessing.
2. Who built it, the lab behind the model.
3. Its context window, how much conversation and source material it can hold at once.
4. What it costs to run, in credits, before you send anything.

The cost line deserves emphasis because most AI subscriptions hide it. On Perspective AI the price of a request is visible before you make it, in stable, dollar-legible credits, whatever the model charges underneath. There is no surprise at the end of the month, just a balance that moves when you use it.

The context window line matters more than most people expect. It is the difference between a model that can hold your whole document and one that quietly forgets the first half, and it varies widely across the catalog. Having it printed on every card means you can pick a long-context model for the contract review and a fast, short-context one for the quick rewrite, deliberately, instead of finding out the limit the hard way. For a deeper treatment of why this number shapes model choice, see our guide to AI context windows.

Routing modes: who picks the model

You decide how much of the model choice to hand off:

Additional modes that optimize for reasoning depth, research, or credit efficiency are on the roadmap. They are not live today, and this page will say so when that changes.

Routing is only half of the multi-model workflow. The other half is movement: you can switch models mid-conversation without losing context, compare two or more models on the same prompt side by side, or combine several models in one thread so each answer comes from the model best suited to it. That behavior has its own deep dive in switching AI models mid-conversation, and the side-by-side method is covered in our AI model comparison guide.

Tuning each request

Two per-request controls apply across the text catalog:

• Reasoning effort: spend more thinking on hard problems, less on simple ones. You set it per request, not per subscription tier.
• Web search toggle: let the model answer from live web results when the question needs current information, or keep it off when it does not.

Together with the model picker, these controls mean the same subscription behaves like a quick chatbot for small questions and a slow, careful reasoner for hard ones, at a visible cost difference per request. A concrete morning looks like this: web search on and a search-native model for the news question, effort turned up on a reasoning flagship for the tricky architecture decision, then a cheap open-weight model with everything off for reformatting the notes. Three different tools in the single-lab world, three requests in one thread here.

Image and video models in the catalog

For completeness: the catalog includes text-to-image and image-editing models, and text-to-video and image-to-video models, metered per generation. They live in the same interface and the same credit balance as the text models. That is the catalog statement; if your main use case is media generation, check the individual model cards in the app for current options and per-generation costs.

How new models get added

Today the catalog is curated: the team adds and maintains models, prioritizing capability, coverage across labs, and genuine open-weight options. New flagship models are typically live within days of release, which is the practical answer to the question every aggregator gets asked: "will you have the next big model?" The pattern so far is yes, within days, in the same subscription.

The longer-term direction is permissionless listing, where anyone can publish a model and incentives flow to it by measured demand rather than popularity votes. That is a stated direction, not a shipped feature, and this catalog page only describes what you can use today.

What the models cost to run

Model pricing is cost-plus-margin: the platform pays the underlying provider for each request and adds a thin margin on top. There is no per-model markup game where the popular model quietly costs five times its provider rate. Rates track real provider costs and are refreshed as those costs move, so the price you see reflects what a model actually costs to run.

Perspective AI does not compete on being the lowest-cost way to call a model, and says so plainly. It competes on breadth and on freedom from being repriced or locked in: 60+ models, one $14.99/mo subscription, and a visible per-request cost for each of them. For the full comparison of what individual lab subscriptions cost against consolidated access, the AI pricing guide has the current numbers for ChatGPT, Claude, Gemini, and the rest of the market.

The bottom line

The catalog question has a short answer: GPT, Claude, Gemini, Grok, DeepSeek, and 60+ models, covering text, reasoning, image, video, and live search, each with a card that tells you what it is good at and what it costs before you use it. The longer answer, the one this page exists for, is that the catalog is built around movement between models rather than commitment to one. Pick your model or let Auto pick it, tune the reasoning effort, and switch mid-thread when a different model would do the job better.

FAQ