Agentic WeeklyGet The Weekly Verdict, freeSubscribe free

ChatGPT vs Claude: we ran both chat apps on 13 tasks (round 1, November 2026)

On Nov 1, 2026 we gave the claude.ai and chatgpt.com chat apps, on paid plans with default settings, the same 13 tasks with hidden graders (round 1 of 3, one attempt each). Claude passed 12 of 13 and ChatGPT 11 of 13. The 95% interval on the gap includes zero, so it's a tie. These are the chat apps, not Claude Code or the Codex CLI. Pick ChatGPT if you want drafts, decks and meeting notes in one place; pick Claude for careful, long-form drafting. Plans and prices differ and are not like-for-like, so we do not name a cheaper tool. Everything except the We ran it section is Reported from vendor pages.

Our pick: On our 13-task round 1 (Nov 1, 2026) it is a tie: Claude 12/13, ChatGPT 11/13, inside the margin of error. Pick ChatGPT for drafts, decks and meeting notes in one place; pick Claude for careful, long-form drafting. Updated (America/Chicago).

By Agentic Weekly, operated independently from Texas, US · Published · Last updated (America/Chicago) · How we work

Evidence: one section is Tested (Nov 1, 2026): the We ran it section, where we ran both chat apps on 13 tasks (round 1 of 3) and an earlier two-task pilot. Everything else is Reported: read from each vendor's own pages and the public GitHub API on 2026-10-25 (America/Chicago), prices checked Nov 1, 2026.

ChatGPT

OpenAI

Reported

Still the broadest all-round assistant; the October updates matter most if you write documents, decks or work with audio.

Claude

Anthropic

Reported

A strong writing partner; try Claude Duet only if you are on Pro or Max and comfortable with a research-preview product.

Side by side, from what we recorded

Each row compares one recorded fact for both tools. Rows are not weighted and nothing is added up into a score; the note under each row says what the marker means.

  • Evidence

    Reported for ChatGPT
    Reported for Claude

    Tested = we ran it (smoke test); Reported = we read official sources

  • Editorial rank: writing

    #1 (ChatGPT is ahead on this row) for ChatGPT
    #2 for Claude

    Lower is higher on our list; an editorial pick, not a benchmark

  • Release date

    Oct 20, 2026 for ChatGPT
    Oct 21, 2026 (Claude is ahead on this row) for Claude

    Newer date only; says nothing about quality

  • Pricing we could verify

    Verified from official page for ChatGPT
    Verified from official page for Claude

    About our evidence only, not about the price

  • Steps in our setup guide

    3 for ChatGPT
    3 for Claude

    Counts our guide's steps, not difficulty

  • Sources we cite

    2 for ChatGPT
    3 for Claude

    How much we had to read; more is not better

Recorded differences, not a quality score; any head-to-head results we ran are in their own section below. How we label evidence.

We ran it: how did ChatGPT and Claude do on 13 tasks?

Tested (Nov 1, 2026), round 1 of 3. Only this section is Tested; the rest of the page is Reported. These are tests of the chat apps (claude.ai and chatgpt.com), not of Claude Code, the Codex CLI, Cursor or the APIs.

On Nov 1, 2026, 2:41 to 3:14 PM (America/Chicago), we pasted the same 13 prompts into each chat app, one fresh chat per task, one attempt each, with default settings. Hidden graders check each answer with edge cases and thousands of random inputs compared against our reference solution. A task passes only if every check passes. How the suite works: methodology.

  • ChatGPT: chatgpt.com, Pro plan as shown. Model picker label GPT-6, Thinking effort Medium (the default). Temporary chat set to Unpersonalized, so memory and custom instructions were ignored.
  • Claude: claude.ai, Pro plan as shown. Model picker label Sonnet 5.5, Medium effort (the default). Incognito chat.
  • Neither app showed tool use (no code runs or web searches) on any task.
Round 1 results, Nov 1, 2026 (13 tasks, one attempt each)
TaskChatGPTClaude
T01 version-range parser (Python, long spec)Pass (second send; code re-copied via Copy code)Pass
T02 find and fix four bugs (Python)Pass (code re-copied via Copy code)Pass
T03 strict CSV parser (JavaScript)PassPass
T04 gaps-and-islands streaks (SQL)PassPass
T05 behavior-preserving refactor (Python)Miss: 5 of 6 checks; one function nests 3 deep, the prompt allows 2 (code re-copied via Copy code)Pass
T06 interval scheduling (Python)Pass (code re-copied via Copy code)Pass
T07 MRR data question (exact answers)PassPass
T08 logic puzzle (exact answer)PassPass
T09 async mapLimit (JavaScript)Pass (second send)Pass
T10 business-hours dates (Python)Pass (code re-copied via Copy code)Pass
T11 release notes (writing, 15 automatic checks)PassPass
T12 instruction following (11 verifiable rules)PassMiss: 10 of 11; "desk" appears 2 times, the prompt required 3
T13 predict a Python program's outputMiss: 9 of 11; line 6 wrong, line 5 has one extra space (see note)Pass
Tasks passed11 of 1312 of 13

Verdict for round 1: a tie. The difference in pass rate (ChatGPT minus Claude) is -0.077, and the bootstrap 95% interval over tasks is -0.308 to 0.154. That interval includes zero, so one task's difference is not a winner. Mean partial score: Claude 0.993, ChatGPT 0.973. Median time per reply was 17 seconds for ChatGPT and 27.5 for Claude, but times were taken by polling the page and are mostly upper bounds, so they are not a speed comparison. Rounds 2 and 3 repeat every task in fresh chats.

Disclosures. ChatGPT T01 and T09 are second sends: on T01 the app's Copy button garbled the first reply and the temporary chat could not be reopened; on T09 the first reply was lost when the tab navigated away before capture (our error). Neither first reply was read or used. Claude needed no second sends. For ChatGPT T01, T02, T05, T06 and T10, the whole-reply Copy button returned truncated code without line breaks, so we re-copied the code with the code block's own Copy code button and merged it with the reply's prose; the raw whole-reply copies are published next to each capture. T11 counts its 15 automatic checks only; its blind quality rubric is not scored yet.

The T13 whitespace question. ChatGPT's line 5 has 4 spaces where Python prints 3. We checked whether that could be a copy artifact. We can't prove it either way, because the temporary chat can't be reopened. But the capture contains no tabs or non-breaking spaces, the copy problems we saw elsewhere were truncated code rather than added spaces, and Claude's copy of the same line has the correct 3 spaces. The result doesn't depend on it: with line 5 corrected, ChatGPT's T13 still fails on line 6 (it printed [2] where Python prints []), 10 of 11 checks. We count it as a miss either way.

Raw data for round 1: README with the disclosures, results.csv, results.jsonl, summary, methodology and the grading harness. Each task's verbatim prompt, grader and reference solution are under /raw/chatgpt-vs-claude-round-1/tasks/ (for example T13 prompt and T13 grader), and every raw reply with its timing is under /raw/chatgpt-vs-claude-round-1/runs/ (for example ChatGPT T13 reply). Re-run the grading with python3 harness/run_all.py runs results. Our test repository: github.com/agentic-weekly/agentic-weekly.

Earlier pilot (Nov 1, 2026, about 1:52 to 1:57 PM): before round 1 we ran a two-task pilot in the same apps, one run each. Both answers passed all our checks on a Python ISO 8601 duration parser (18 of 18) and a JavaScript median fix (7 of 7), a tie on easy tasks, which is why the 13-task suite is harder. Pilot prompts, outputs and grading scripts on GitHub. Looking for the terminal coding agents? We have not run those yet: Claude Code vs Codex CLI.

What else have we measured, and what have we not?

Apart from the round 1 tasks and the pilot in the We ran it section above, we have not run ChatGPT or Claude ourselves, so the rest of this page is Reported from vendor and GitHub pages. Below is what we still plan to run.

Our own tests for ChatGPT and Claude
TestChatGPTClaude
Hands-on test log (install and a real task)Not tested yet. What we will run: our writing and instruction-following tests (same prompts, blind-rated, code-checked).Not tested yet. What we will run: our writing and instruction-following tests (same prompts, blind-rated, code-checked).
Writing test (same prompts, blind-rated by a separate model)Not tested yet. We have not run our writing and instruction-following tests on this product. We have run them on open models only: writing leaderboard, instruction following.Not tested yet. We have not run our writing and instruction-following tests on this product. We have run them on open models only: writing leaderboard, instruction following.

How we test and label results: methodology. All test logs: test hub.

How do ChatGPT and Claude compare at a glance?

ChatGPT and Claude are both listed under Writing. Agentic Weekly's ChatGPT page names Claude as an alternative. ChatGPT: Still the broadest all-round assistant; the October updates matter most if you write documents, decks or work with audio. Claude: A strong writing partner; try Claude Duet only if you are on Pro or Max and comfortable with a research-preview product.

ChatGPT and Claude side by side (vendor pages and GitHub, 2026-10-25)
FeatureChatGPTClaude
VendorOpenAIAnthropic
Listed underWriting, Image generation, Small businessWriting
LicenseProprietary hosted productProprietary hosted product
Pricing modelFree tier + paid plans (per its pricing note)Free tier + paid plans (per its pricing note)
Free tierAppears in the plan list in the Paid plans rowAppears in the plan list in the Paid plans row
Paid plansPer chatgpt.com/pricing and OpenAI's pricing docs (checked Nov 1, 2026): Free $0, Go $8/month (the page says it may include ads), Plus $20/month, Pro plans at $100, $200 or $500/month, Business $20 per user per month billed annually ($25 billed monthly, 2+ users), plus Enterprise and Edu on request.Per Anthropic's pricing page (checked Nov 1, 2026): Free $0; Pro $20/month or $17/month billed annually ($200 upfront); Max $100/month (5x Pro usage); Ultra $200/month (formerly Max 20x; 10x Pro weekly, no five-hour limit for verified accounts, renamed Oct 30, 2026); Team Standard $20 or $25 per seat and Premium $100 or $125 per seat (annual or monthly); Enterprise $20 per seat per month billed annually with a 20-seat minimum, with usage on top.
EvidenceReported from vendor pagesReported from vendor pages

How do ChatGPT and Claude differ?

  • Plans and prices differ and are not like-for-like (credits, seats and usage limits vary), so we do not name a cheaper tool. The Paid plans row above quotes each vendor's plan list; the price source line near the top of this page links each vendor's pricing page and says when we last checked it.
  • ChatGPT is "Free tier + paid plans (per its pricing note)"; the entry for Claude has no pricing model we could confirm (see its pricing note).

Neither tool publishes open-source code under a standard license, so you are choosing between hosted products.

Which should you pick, ChatGPT or Claude?

  • Pick ChatGPT if: writers who need drafts, decks and meeting notes in one place; teams already on ChatGPT Business or Enterprise. Skip it if: you want strict data isolation by default (check settings).
  • Pick Claude if: writers who want a careful, long-form drafting assistant; pro and Max users curious about proactive agents. Skip it if: you need Duet and are on Free (it is Pro and Max only).

What are the strengths and limits of each?

Strengths and limits as listed on each tool page
FeatureChatGPTClaude
StrengthsVery broad feature set; Detailed dated release notes; Word, Excel and PowerPoint add-insDated, short release notes; Haiku 5.5 targets cost-sensitive use; Monthly API credits for Max and Team
LimitsMany features roll out gradually and vary by plan, region and workspace; Models, plans and limits change often; Custom GPTs are planned to be retired in favor of pluginsDuet is a research preview and limited to Pro and Max; Release notes are brief; product details are in separate announcements; We did not test writing quality

What did we not verify?

  • ChatGPT: Writing quality compared with other assistants (not tested).
  • Claude: How well Claude Duet works (preview by waitlist; not tested).
  • Output quality, speed and reliability beyond the 13 round 1 tasks and the pilot in the We ran it section: not compared.

What others found

ChatGPT vs. Claude: Which Should You Use in 2026? · The AI Advantage ·
Watch on YouTube

Third-party video, not one of our tests. Nothing here is marked Tested because of it.

FAQ

Is ChatGPT or Claude better at coding?

On our 13-task round 1 (Nov 1, 2026), run in the chatgpt.com and claude.ai chat apps on paid plans, Claude passed 12 tasks and ChatGPT 11. The 95% interval on the gap includes zero, so it is a tie. That tests the chat apps, not Claude Code or the Codex CLI, and it is round 1 of 3.

Is ChatGPT or Claude open source?

ChatGPT: Proprietary hosted product. Claude: Proprietary hosted product. Licenses come from each project's GitHub repository or the vendor's own pages, read Oct 25, 2026; read the license text itself before redistributing or building a competing service.

Does ChatGPT or Claude have a free plan?

ChatGPT: its plan list is quoted in the Paid plans row above. Claude: its plan list is quoted in the Paid plans row above. Checked Nov 1, 2026; prices and limits change, so confirm on each vendor's page.

What is the main difference between ChatGPT and Claude?

Plans and prices differ and are not like-for-like (credits, seats and usage limits vary), so we do not name a cheaper tool. The Paid plans row above quotes each vendor's plan list; the price source line near the top of this page links each vendor's pricing page and says when we last checked it.

Sources

Page history

  • : First published. Facts read from vendor pages and the GitHub API on 2026-10-25.
  • : Added the We ran it section: both chat apps on two coding tasks (Tested, Nov 1, 2026). Rest of the page unchanged and still Reported. Now a hand-maintained page.
  • : Added both test prompts verbatim to the We ran it section.
  • : Published round 1 of 3 of our 13-task chat-app suite (Claude 12/13, ChatGPT 11/13, a tie) with raw data and disclosures. The two-task pilot stays on the page, labelled as a pilot.

Markdown version for agents: /compare/chatgpt-vs-claude.md