You are choosing one chatbot subscription and want to know what $20 to $30 a month actually buys. ChatGPT, Claude, and Grok are the consumer assistants from OpenAI, Anthropic, and xAI. This page lays out plans, message caps, the model on each tier, headline benchmarks, and what each company does with your conversations, every number sourced.
The Short Answer
ChatGPT Plus and Claude Pro both cost $20/month. SuperGrok costs $30/month. All three have a free tier. The cheapest paid plan anywhere is ChatGPT Go at $8/month (ad-supported, available in 98 countries since January 2026), and Claude Pro falls to $17/month if you pay annually.
The models differ. ChatGPT runs GPT-5.5. Claude runs Claude Fable 5, released June 9, 2026, Anthropic's most capable widely released model (currently suspended, see note above). Grok runs Grok 4.3, xAI's latest, with a December 2025 knowledge cutoff and built-in real-time search from X. Pick on which model and which limits fit how you work, not on price alone, because the two paid leaders are tied at $20.
Benchmarks are self-reported
Each vendor publishes its own numbers using its own test harness. Treat cross-vendor benchmark comparisons as directional, not exact. Prices, message caps, and data-training defaults are documented on each vendor's own pages and are linked in the Sources section.
Plans and Pricing
| Plan level | ChatGPT | Claude | Grok |
|---|---|---|---|
| Free | $0 (ads in US) | $0 | $0 |
| Cheapest paid | Go: $8/mo (ad-supported) | Pro: $17/mo annual | SuperGrok: $30/mo |
| Standard paid | Plus: $20/mo | Pro: $20/mo monthly | SuperGrok: $30/mo |
| Power tier | Pro: $100/mo, Pro Max: $200/mo | Max 5x: from $100/mo, Max 20x | SuperGrok Heavy |
ChatGPT Pro at $100/month gives 5x Plus usage, elevated Codex limits, and 125 Deep Research queries per month. Pro Max at $200/month adds 20x usage, GPT-5 Pro, and unlimited audio and video mode. Claude Max 5x starts at $100/month for 5x Pro usage; Max 20x gives 20x; Max plans are monthly-only with no annual discount. SuperGrok Heavy adds Grok 4 Heavy, the highest usage limits, and 16x AI agents on Expert mode.
For teams: ChatGPT Business is $20/user/month annual or $25/user/month monthly (minimum 2 users). Claude Team runs 5 to 150 people at $20/seat annual ($25 monthly) for the Standard seat, with a $100/seat Premium tier.
Free Tier and Rate Limits
The limit you hit day to day matters more than the sticker price. Each app throttles differently.
| ChatGPT | Claude | Grok | |
|---|---|---|---|
| Free-tier cap | 10 GPT-5.5 messages per 5 hours, then mini | Compute-based, then smaller models | Compute-based, then smaller models |
| Paid message cap | Plus/Go: 160 GPT-5.5 messages per 3 hours | Weekly limit across all models, 5-hour session limit | Higher limits on SuperGrok |
| Reasoning model cap | GPT-5.5 Thinking: up to 3,000 messages/week on Plus | Max: separate weekly Sonnet limit | Highest limits on SuperGrok Heavy |
| On hitting the cap | Switches to mini model until reset | Resets 7 days after session start | Shifts to smaller model |
ChatGPT publishes hard message counts: Free gets 10 GPT-5.5 messages every 5 hours; Plus and Go get up to 160 every 3 hours, then drop to the mini model; GPT-5.5 Thinking can be selected manually with up to 3,000 messages per week on Plus or Business. Claude uses a weekly usage limit that applies across all models and resets seven days after the session starts, plus a five-hour session limit visible in Settings, Usage; Max plans add a second weekly limit for Sonnet only. Grok caps by usage tier, with SuperGrok and SuperGrok Heavy unlocking progressively higher limits.
Which Model Runs Where
| ChatGPT | Claude | Grok | |
|---|---|---|---|
| Flagship | GPT-5.5 | Claude Fable 5 | Grok 4.3 |
| Reasoning variant | GPT-5.5 Thinking, GPT-5 Pro | Opus, Sonnet, Haiku | Grok 4 Heavy (Heavy tier) |
| Knowledge / notes | GPT-5.5 and 5.5 Pro in API since Apr 24, 2026 | Fable 5 = Mythos 5 with extra safeguards | Dec 2025 cutoff, real-time X search |
GPT-5.5 is the latest GPT model live in ChatGPT; GPT-5.5 and GPT-5.5 Pro reached the API on April 24, 2026. Claude Fable 5 and Claude Mythos 5 became available June 9, 2026; Fable 5 is built for demanding reasoning and long-horizon agentic work and is the same underlying model as Mythos 5 with added cybersecurity and biology safeguards (queries in those domains route to Opus 4.8). Claude also exposes Opus, Sonnet, and Haiku, with availability varying by tier. Grok 4.3 is xAI's latest and fastest model, with an improved architecture, a December 2025 knowledge cutoff, native tool use, and real-time search; from Grok 4.3 it can also write and run code, install dependencies, and produce real files.
Benchmarks
Published, vendor-reported scores. Different harnesses, so read these as directional.
| Benchmark | ChatGPT | Claude | Grok |
|---|---|---|---|
| SWE-Bench Pro (coding) | 58.6% (GPT-5.5) | n/a | n/a |
| SWE-bench Verified (coding) | n/a | 80.9% (Opus 4.5) | n/a |
| Terminal-Bench 2.0 | 82.7% (GPT-5.5) | n/a | n/a |
| GPQA Diamond (science) | 93.2% (GPT-5.2 Pro) | 87.0% (Opus 4.5) | n/a |
| Humanity's Last Exam | n/a | n/a | 50.7% (Grok 4 Heavy, text-only) |
| ARC-AGI-2 (verified) | 54.2% (GPT-5.2 Pro) | n/a | n/a |
| LMArena Text Arena | n/a | n/a | 1483 Elo, #1 (Grok 4.1 Thinking) |
GPT-5.5 reaches 58.6% on SWE-Bench Pro and 82.7% on Terminal-Bench 2.0; GPT-5.2 Pro scores 93.2% on GPQA Diamond and 54.2% on ARC-AGI-2 (Verified). Claude Opus 4.5 scores 80.9% on SWE-bench Verified and 87.0% on GPQA Diamond (averaged over 5 trials, 64k thinking budget, 200k context, high effort). Grok 4 Heavy was the first model to hit 50.7% on Humanity's Last Exam (text-only subset), and Grok 4.1 Thinking holds the #1 overall spot on LMArena Text Arena at 1483 Elo, 31 points above the highest non-xAI model. Where a cell reads "n/a," that vendor has not published a comparable number for that test.
Features
ChatGPT (GPT-5.5)
Plus includes ChatGPT Agent, Codex Agent, a Sora 1 video preview, Canvas, Advanced Data Analysis, Deep Research, Vision, and custom GPTs. Business adds SSO, MFA, admin controls, and connectors to Google Drive, SharePoint, GitHub, and Notion.
Claude (Fable 5)
Free includes chat across web and mobile, code generation, web search, memory, file creation, and connectors. Pro adds Claude Code, Claude Cowork, unlimited projects, Research, multiple models, and Microsoft 365/Outlook integration.
Grok (4.3)
Real-time search from X, native tool use, and fewer guardrails. From Grok 4.3 it can run code, install dependencies, and produce files; Skills apply persistent expertise across conversations on web, iOS, and Android.
Your Data and Training
The defaults differ, and the difference is worth one setting toggle.
| ChatGPT | Claude | Grok | |
|---|---|---|---|
| Trains on chats by default? | Yes (consumer) | User choice on Free/Pro/Max | Yes, with opt-out |
| How to opt out | Settings, Data Controls, off | Model-improvement setting off | Account privacy settings |
| Retention if opted out | History kept, not used to train | 30-day retention | Varies |
ChatGPT's "Improve the model for everyone" toggle is on by default for consumer accounts; switching it off in Settings, Data Controls stops new conversations from being used to train models. The setting applies account-wide regardless of device, can be changed anytime, and chats still appear in history without being used for training. On Claude, model improvement is a user choice for Free, Pro, and Max; with it off, Anthropic does not use previous or new chats for training and applies a 30-day retention, while allowing chats to improve Claude permits de-identified retention of up to 5 years in training pipelines.
Pick by Use Case
| You want | Best fit | Why |
|---|---|---|
| Cheapest paid plan | ChatGPT Go ($8/mo) | Lowest price; ad-supported; 98 countries. |
| Latest, most capable model | Claude (Fable 5, currently suspended, see note) | Anthropic's most capable widely released model. |
| Breaking news and real-time | Grok (4.3) | Native real-time search from X. |
| Math and science reasoning | ChatGPT | 93.2% GPQA Diamond, top ARC-AGI-2. |
| Coding agent in the terminal | Claude | Claude Code bundled on Pro and above. |
| Privacy by default | Claude | Training is a user choice; 30-day retention if off. |
| Video generation | ChatGPT | Sora 1 preview on Plus. |
Frequently Asked Questions
Which is cheapest: ChatGPT, Claude, or Grok?
ChatGPT Plus and Claude Pro are both $20/month; SuperGrok is $30/month. Cheaper options: ChatGPT Go at $8/month and Claude Pro at $17/month billed annually. All three have a free tier.
What model does each chatbot run in June 2026?
ChatGPT runs GPT-5.5, Claude runs Claude Fable 5 (released June 9, 2026), and Grok runs Grok 4.3 (December 2025 knowledge cutoff). Claude also offers Opus, Sonnet, and Haiku; ChatGPT offers GPT-5.5 Thinking and GPT-5 Pro on higher tiers.
How many free messages do you get?
ChatGPT Free allows up to 10 GPT-5.5 messages every 5 hours before dropping to a mini model. Claude and Grok cap the free tier by a rolling compute-based limit rather than a fixed message count, then shift to smaller models.
Which chatbot trains on your conversations by default?
ChatGPT trains on consumer chats by default, with an off switch in Settings, Data Controls. Claude makes training a user choice for Free, Pro, and Max; with it off, Anthropic does not train on your chats and keeps a 30-day retention.
Which is best for coding?
On published numbers, GPT-5.5 hits 58.6% on SWE-Bench Pro and Claude Opus 4.5 hits 80.9% on SWE-bench Verified. Claude bundles Claude Code, a terminal agent, on Pro and above. For sustained engineering, pair a frontier model with a dedicated coding agent.
Related comparisons
ChatGPT vs Grok
GPT-5.5 vs Grok: real-time X data and fewer guardrails vs ecosystem maturity.
ChatGPT vs Claude vs Gemini
The three frontier assistants compared on coding, writing, multimodal, and price.
Claude vs ChatGPT
An honest comparison from a team that routes production traffic to both.
ChatGPT vs Gemini
GPT-5.5 vs Gemini 3.1 Pro: benchmarks, pricing, multimodal, and when to route to each.
ChatGPT vs Microsoft Copilot
Copilot runs on OpenAI models inside Microsoft 365. Where the wrapper helps and where it taxes you.
ChatGPT vs Perplexity
General assistant vs answer engine: cited search vs open-ended generation.
Sources
Building With More Than One Model? Route Across All Three.
If you call these models through an API instead of the chat box, an LLM router can send each request to the right model automatically: hard reasoning to one, code to another, high-volume to the cheapest. Classify difficulty per request across OpenAI, Anthropic, and xAI.