This is the fastest-moving product category on the internet, and comparison articles rot quickly. Before getting to recommendations, it is worth correcting four claims that are still repeated constantly — including, until this update, on this page.
Similarly, Claude's often-quoted "200K token context window" is now the floor rather than the ceiling: Anthropic documents 1 million tokens on Opus 5 and Sonnet 5 for paid plans, 500K on several 4.x models, and 200K for models outside those groups.
ChatGPT organises itself around three pillars its own site names as chat, work and code. The documented feature surface includes ChatGPT Work, the Codex coding agent, Deep Research, Study Mode, voice and voice-with-video, Sites, Projects, scheduled tasks, custom GPTs and a Skills beta. The model family is GPT-5.6, with named variants including Sol, Sol Pro, Terra and Luna.
Plans: Free, Go, Plus and higher tiers. Free tier: yes. Reasonable pick if you want the broadest set of capabilities under one subscription and value the largest third-party ecosystem. Current plans: chatgpt.com/pricing.
Claude's documented surface includes Artifacts, Projects, Research, Skills, Connectors (remote MCP support), Claude Code, Claude Cowork, Claude Design, Claude Science, a Claude Security beta, plus Chrome and Microsoft 365 integrations. The model lineup spans Fable, Opus, Sonnet and Haiku tiers.
Free tier: yes. Context: up to 1M tokens on Opus 5 and Sonnet 5 for paid plans — the most generous documented limit among these four, which matters if you routinely work across large documents or codebases. Current plans: claude.com/pricing.
Gemini's current app models are the Gemini 3 generation — Flash-Lite, Flash and Pro tiers, with point releases through 2026 including Gemini 3.1 Pro and Gemini 3.6 Flash, plus a Deep Think reasoning mode. Documented features include Deep Research, Gems (custom assistants), Canvas, Audio Overviews, Storybook, Gemini Notebook, the Nano Banana Pro image generation and editing line, and Gemini Spark, a persistent personal agent limited to higher tiers and select countries.
Plans: Google AI Plus, Google AI Pro, Google AI Ultra. Free tier: yes, with access to all three current app models at standard limits. Reasonable pick if you live in Google Workspace, or you want a very large context window without paying a premium tier elsewhere. Current plans: Google AI plans.
Perplexity is structurally different from the other three: it is a model-orchestration product rather than a single model. Its own documentation states that it routes queries across multiple frontier models — naming ChatGPT, Gemini, Claude and NVIDIA Nemotron — and selects per query. Documented features include Comet, its agentic browser; Perplexity Computer, which runs multiple models as sub-agents on complex tasks; Deep Research, which can now produce deliverables such as slides and sheets; Model Council, which runs a query across several models in parallel and synthesises the answers; and Brain, a memory system in research preview.
Plans: Free, Perplexity Pro, Perplexity Max. Free tier: yes — search with citations and access to basic models. Notable 2026 change: Comet came to iOS and dropped its paywall in March 2026. Reasonable pick if your work is research-shaped and source citation matters more than raw generation. Current plans: perplexity.ai/hub/pricing.
| Dimension | ChatGPT | Claude | Gemini | Perplexity |
|---|---|---|---|---|
| Free plan | Yes | Yes | Yes | Yes |
| Paid tier names | Go, Plus, higher | Pro, higher | AI Plus / AI Pro / AI Ultra | Pro, Max |
| Current model family | GPT-5.6 | Fable / Opus / Sonnet / Haiku | Gemini 3 generation | Routes to third-party models |
| Max documented context | See vendor docs | 1M tokens (Opus 5 / Sonnet 5, paid) | 1M tokens (AI Pro / Ultra) | Varies by routed model |
| Real-time web access | Yes | Yes | Yes | Yes — core to the product |
| Agentic browser | — | Chrome integration | — | Comet |
| Coding agent | Codex | Claude Code | — | — |
This table records what vendors document, not how well any of it performs. Where a cell says "see vendor docs," it is because we could not verify a current figure from a primary source at the time of writing, and we would rather leave a gap than guess.
Monthly fees are the visible cost of these subscriptions and, at the entry paid tier, they are close enough between the four that price is rarely the deciding factor. The cost that actually accumulates is switching cost, and it has grown sharply as each vendor has added persistent state.
Consider what you leave behind when you move. On ChatGPT that is custom GPTs, Projects, saved memory and scheduled tasks. On Claude it is Projects, Skills and any Connectors you have wired into your own systems. On Gemini it is Gems and whatever workflow you have built around Workspace integration. On Perplexity it is Spaces and, if you are on the higher tier, whatever Brain has accumulated about how you work. None of this is portable between vendors, and none of it shows up on a pricing page.
The practical implication is that the decision matters more than the monthly figure suggests, and that it is worth spending a few weeks on free tiers before you start building persistent structure inside any one of them. It is also an argument for keeping your genuinely important reference material in your own files rather than inside an assistant's memory, regardless of which one you choose.
Because all four document a free plan, the free tiers deserve more attention than most comparisons give them. They differ in kind, not just in generosity.
Perplexity's free tier is the least compromised relative to its paid product, because the core job — search with citations — is available without paying; the paid tiers mainly add model choice, Model Council and Computer. Gemini's free tier gives access to all three current app models but caps context at 32K tokens, which is the single biggest functional difference between free and paid in this group and will bite quickly if you work with long documents. Claude and ChatGPT both gate their strongest models and heaviest features behind paid plans, so their free tiers are best understood as evaluation tools rather than as products you settle into.
If you are choosing on free tiers alone, the honest ranking is by fit rather than quality: Perplexity for research, Gemini for anything already inside Workspace, and Claude or ChatGPT as trials rather than destinations.
Because all four have a free tier, the sensible sequence is to try before subscribing — and to notice that the interesting differences in 2026 are no longer about raw answer quality, which has converged, but about the surrounding machinery. Claude and ChatGPT both ship coding agents. Perplexity ships a browser. Gemini ships deep Workspace integration and a persistent agent on higher tiers.
A reasonable decision rule: pick Perplexity if most of your questions are research questions needing citations; pick Gemini if your documents already live in Google Workspace; pick Claude if you work across very large documents or codebases and want the largest documented context window; pick ChatGPT if you want the widest range of tasks and integrations in one place. Many people end up with two, and given free tiers exist, running one paid subscription alongside a second free account is a common and cheap arrangement.
Related reading: our AI writing tools comparison, our roundup of the best free AI tools, and the AI tools guide.
No. Per OpenAI's help documentation the free default sits in the GPT-5.6 family, and a lower-cost Go tier now sits between Free and Plus.
Yes, with citations, across plans. Comparisons claiming otherwise are out of date.
No. That is a model name. The plans are Google AI Plus, Google AI Pro and Google AI Ultra.
Model-dependent: 1M tokens on Opus 5 and Sonnet 5 for paid plans, 500K on several 4.x models, 200K otherwise.
All four, per their own pricing pages.
No, and we no longer imply that we do. This page is compiled from vendor documentation, dated at the top.
We update this page when the vendors change something that matters. Get a note when we do.