Claude vs ChatGPT: models, pricing, and coding

Claude and ChatGPT overlap on chat, research, files, and code. No universal intelligence ranking settles which one to buy. Your task mix settles it, along with the interfaces you actually open and the usage limits you actually hit. This comparison uses the vendors' live pricing pages and treats forum sentiment as a source of test cases, not evidence.

By the Continuum team. We build a workbench that runs Claude Code, Codex, and their peers, so the model rates quoted here are the ones our own cost analytics ship with.

The short version

Buy Claude when Claude Code, the current Claude lineup, and Claude's document and research behavior carry your day. Buy ChatGPT when the wider ChatGPT tool set and Codex carry it instead. Claude Pro and ChatGPT Plus both list at $20 a month, so if both look plausible, pay for one month of each and run the same real tasks before you commit to annual billing.

What you need to know
  • Claude Pro is $20 monthly, or $17 a month billed annually. ChatGPT Plus is $20 monthly. Same headline price, different limits.
  • The step up costs the same on both sides: Claude Max starts at $100 a month, ChatGPT Pro is $100 for 5x Plus usage and $200 for 20x.
  • Claude's August 2026 lineup is Fable 5, Opus 5, Sonnet 5, and Haiku 4.5. ChatGPT and Codex run the GPT-5.6 family.
  • For developers the real comparison is Claude Code against OpenAI Codex, not one chat reply against another.
  • Vendor model descriptions are positioning. They prove nothing about your repository.
  • Reddit is good for finding test cases. It is worthless for current pricing, limits, or model behavior.

Claude vs ChatGPT in one decision table

Search engines file this question under both Claude vs ChatGPT and Claude AI vs ChatGPT. Same question, same answer: there is no responsible one-line winner. The two vendors change models and plan limits independently, and each describes its own models in its own evaluation language. Compare the whole workflow instead: conversation quality, the coding agent, files and research, usage headroom, and which interfaces exist on the devices you use.

Start with the workflow, then test quality

If your priority isEvaluate firstWhy
Repository-scale codingClaude Code and Codex on the same issueAgent behavior, tool use, review quality, and recovery matter more than a chat answer
Fast everyday questionsClaude Sonnet and GPT-5.6 LunaLatency and concise instruction-following dominate routine work
Hard reasoningClaude's higher-capability models and ChatGPT Sol or Sol ProBoth vendors reserve different capabilities and limits for higher tiers
Research and filesThe same source packet in both productsCitation quality, file limits, and synthesis style are task-specific
One subscription for mixed personal workThe complete plan feature listVoice, image, connectors, projects, memory, and coding access differ

A fair test uses the same inputs and an objective acceptance check. For code, use the same commit, issue text, test command, and time limit. For research, require source links and verify every cited claim. For writing, blind-review the drafts without product names.

Claude vs ChatGPT pricing in August 2026

These are the subscription prices on the vendors' live pages as of this update. They exclude API usage, taxes, regional currency conversion, promotional credits, and temporary capacity changes.

Individual subscription tiers as of August 8, 2026

ProductPlanCurrent listed priceDeveloper-relevant note
ClaudeFree$0The free plan covers Claude chat; Pro explicitly includes Claude Code
ClaudePro$20 monthly, or $17/month billed annually at $200 upfrontIncludes Claude Code and more usage than Free
ClaudeMaxFrom $100/monthOffers 5x or 20x the usage allowance of Pro
ChatGPTFree$0Includes limited access to current ChatGPT features and Codex
ChatGPTGoLocalized monthly price shown at checkoutA lower-cost paid tier whose listed price varies by market
ChatGPTPlus$20/monthAdds higher limits and broader access to models, tools, and Codex
ChatGPTPro$100/month for 5x Plus usage or $200/month for 20x Plus usageHigher individual limits and access to the top reasoning tier

The matching $20 price makes a one-month bake-off cheap. It does not make the usage limits equivalent. Limits move with model choice, request size, tool use, and current plan policy, so read the live plan page and watch your own consumption before you commit to annual billing.

Current Claude and ChatGPT models

Anthropic's pricing page puts Fable 5 on long-running agents, Opus 5 on complex agentic coding and enterprise work, Sonnet 5 on coding and agents, and Haiku 4.5 at the fast, cheap end. Which ones you actually get, and how much you can use them, depends on your plan and the surface you are in.

OpenAI's pricing page lists GPT-5.6 Luna for everyday text chat, GPT-5.6 Sol for advanced reasoning, and Sol Pro as the Pro reasoning tier. Its Work and Codex surfaces add GPT-5.6 Terra alongside Sol and Luna, again by plan.

How the vendors position their current model families

Use caseClaude options to testChatGPT or Codex options to test
Everyday chat and quick editsSonnet 5 or Haiku 4.5GPT-5.6 Luna
Long-running or complex agent workFable 5 or Opus 5GPT-5.6 Sol or Sol Pro
Agentic codingOpus 5 or Sonnet 5 through Claude CodeGPT-5.6 family through Codex
Everyday or lighter tasksHaiku 4.5 or Sonnet 5Luna, or Terra where available in Codex

Those descriptions are vendor positioning, not a cross-vendor benchmark. Model names also do not guarantee identical behavior across chat, desktop, command-line, IDE, and cloud surfaces. Record the exact model, plan, and interface when you compare results.

Strengths by task type

Skip the question of which assistant is smarter and map each task to an observable outcome. The same person can prefer Claude for one category and ChatGPT for another without contradicting themselves.

TaskWhat to test in ClaudeWhat to test in ChatGPTAcceptance check
Long document synthesisInstruction retention across the full source packetResearch and file workflow across the same packetEvery conclusion traces to a supplied source
Daily writingTone control and sustained structureDrafting plus the surrounding ChatGPT tool setBlind review against a written rubric
DebuggingClaude Code's repository search, edits, tests, and explanationCodex's repository search, edits, tests, and explanationThe same regression test passes without unrelated changes
ArchitectureAlternative designs and explicit failure modesAlternative designs and explicit failure modesReviewers can trace tradeoffs to constraints
Data and researchResearch, file analysis, and citationsDeep research, data tools, and citationsClaims survive manual source verification
Multimodal workImage and document understanding available on the chosen planVoice, image, and visual tools available on the chosen planThe output completes your exact end-to-end workflow

Scoring this way stops feature breadth from hiding weak execution. A plan can list twenty tools and still fail the one task you bought it for. It also stops a single impressive answer from hardening into a permanent ranking that survives the next model release.

Claude Code vs the ChatGPT and Codex coding surfaces

For developers the comparison that decides the purchase is Claude Code against OpenAI Codex. ChatGPT is the account and the conversational shell wrapped around OpenAI's coding tools. Both vendors ship coding agents that reach well past a chat box.

Officially documented coding surfaces

SurfaceClaudeOpenAI
TerminalClaude Code CLICodex CLI
EditorClaude Code extensions for VS Code and JetBrainsCodex IDE extension
DesktopClaude Code in the Claude desktop appCodex mode in the ChatGPT desktop app
WebClaude Code on the webCodex web
Background or remote workRemote Claude Code sessions through supported web and app workflowsCodex cloud tasks through supported Codex workflows

Claude Pro and higher individual plans include Claude Code. OpenAI currently includes Codex across ChatGPT plans, with plan-dependent limits and capabilities. Inclusion is only the first question. The decisive differences are how each agent plans, edits, runs commands, handles approval, recovers from errors, explains a diff, and fits your terminal or editor habits.

Use one real repository exercise to compare them. Give both agents a failing test and the same scope constraint. Require a written diagnosis before edits, a minimal patch, the exact test output, and a final diff review. Count manual corrections, unrelated changes, elapsed time, and tokens or usage consumed. Repeat with at least one refactor and one review-only task before choosing a coding subscription.

For the agent-level detail, read Codex vs Claude Code and AI coding pricing compared. Both separate coding-agent economics from the wider Claude and ChatGPT bundles.

How to use Reddit sentiment without treating it as evidence

Reddit tells you which questions developers care about. It cannot verify a current model, price, limit, or feature. Behind every confident post sits an unknown plan, prompt, repository, model selection, date, and tolerance for retries. This guide does not invent quotes or promote anecdotes into a measured consensus.

Treat sentiment as a hypothesis generator. A post says Claude follows a large refactor better: reproduce that refactor with a test suite. Another says ChatGPT is the better general subscription: list the non-coding tools you actually open in a week. Someone says one model is smarter: ask which task, which acceptance check, and which version produced the judgment.

01

Capture the claim

Rewrite the post as a falsifiable statement about one task.

02

Record the setup

Note plan, model, interface, date, repository state, and prompt.

03

Run both products

Use identical inputs and allow the same number of corrections.

04

Score the output

Use tests, source verification, blind review, or another external measure.

Sentiment earns its place when it improves your evaluation set. It fails when popularity stands in for reproducible evidence.

Which subscription should a developer buy?

Start on Claude Pro if Claude Code and Claude's writing, document, and reasoning behavior carry your week. Start on ChatGPT Plus if the wider ChatGPT tool set and the Codex workflow carry it instead. Step up to Max or Pro only after the $20 tier blocks real work more than once. The higher tiers buy headroom and top-tier model access, not correctness.

Paying for both is rational when one implements and the other reviews it. It is waste when both answer the same casual prompts. Track for one month which product finished billable or time-saving work, then cancel the one that did not pay for itself.

This page keeps Continuum out of the matchup. It is neither Claude nor ChatGPT, so it does not belong on either side of the table. Decide between the two vendors on their current products, their current prices, and what they produce in your own workload.

Questions people ask

Neither wins across the board. Both expose several models with different speed, reasoning, tool, and usage profiles, and the lineups change every few months. Run the same real task through both with an objective acceptance check rather than trusting one anecdote or one vendor benchmark.

Claude Pro and ChatGPT Plus both list at $20 per month. Claude Pro is also $17 per month when billed annually at $200 upfront. Both vendors offer free tiers and higher individual tiers, while team and enterprise pricing follows separate plans.

Whichever produces correct, scoped diffs in your repository with fewer corrections. Compare Claude Code against OpenAI Codex rather than the two chat boxes: both agents run in the terminal, an editor, the desktop app, and the web.

Yes. OpenAI currently includes Codex with ChatGPT Plus and other ChatGPT plans, with usage limits and available capabilities that vary by plan.

Yes. Anthropic's current pricing page includes Claude Code with Claude Pro. Max and organizational plans provide their own access and usage allowances.

Anthropic's August 2026 lineup lists Fable 5, Opus 5, Sonnet 5, and Haiku 4.5. Availability and limits depend on the Claude plan and interface.

OpenAI currently lists GPT-5.6 Luna for everyday ChatGPT text, GPT-5.6 Sol or Sol Pro for reasoning on eligible plans, and GPT-5.6 Terra across eligible Work and Codex experiences.

Only when each one does a distinct job, such as one implementing and the other reviewing independently. Run both against the same workload for a month at $40 total, record what each finished, and cancel the one that did not pay for itself.

Sources

Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.

  1. Anthropic: Claude pricing Pro, Max, and what each tier includes
  2. OpenAI: ChatGPT pricing Free, Go, Plus, and Pro tiers
  3. Anthropic: Claude Code platforms the terminal, editor, desktop, and web surfaces
  4. OpenAI: using Codex with your ChatGPT plan which plans include Codex and at what limits
Try it

Decide on evidence,
not brand loyalty.

Run the same tasks through both, score the outcomes, and pay for the workflow that earns its place. Continuum shows what each agent actually cost you while you decide.

free app · your subscriptions · local-first