Which

Claude 3.5 vs GPT-4: Where Those Models Stand Now

The current model lists do not resolve the status of the exact Claude 3.5 and GPT-4 labels, so compare present plans, limits, controls, and cautions instead.

Sources checked 2 Oct 2026

For claude vs chatgpt, the vendors’ current documentation does not settle a like-for-like verdict on the exact labels Claude 3.5 and GPT-4: OpenAI’s model list and Anthropic’s model overview point to newer families, and neither page states whether those older labels are active, legacy, deprecated, or retired. They do document present plans, limits, data controls, and accuracy cautions, but Chat Picker has not tested the assistants for this use.

What the vendors document

Start with the exact label. OpenAI’s current API model list names GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. Anthropic’s model overview names Claude Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 4.5. Both pages were checked on October 1, 2026, but neither states whether “GPT-4” or “Claude 3.5” is active, legacy, deprecated, or retired in the consumer apps.

The deprecation pages, also checked on October 1, 2026, do not fill that gap. OpenAI’s deprecation page says gpt-5.4-cyber is deprecated for API removal on October 1, 2026; whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-transcribe-diarize are scheduled for removal on February 26, 2027; legacy audio, realtime, and transcription families and snapshots are scheduled for January 20, 2027; and older GPT Image models are scheduled for December 1, 2026. OpenAI says generally available models receive at least six months’ notice, specialized generally available variants receive at least three months, and preview models may receive much shorter notice.

Anthropic’s deprecation page says Claude Opus 4.1 retired on August 5, 2026. It also says developers using Claude Sonnet 4.5 were notified on September 30, 2026, of an upcoming Claude API retirement, but it does not give a status or retirement date for Claude 3.5.

The release pages add current availability without resolving the older labels. OpenAI’s release notes list Pro 200 and Pro 500, while Claude’s release notes list a beta 1M-token context window for Sonnet 4.6.

As read on October 1, 2026, OpenAI’s pricing page lists Plus at $20 per month and Pro 100, Pro 200, and Pro 500 at $100, $200, and $500 per month. Anthropic’s pricing page lists Pro at $20 per month on monthly billing or $17 per month with annual billing, and Max from $100 per month with 5x or 20x the Pro usage.

On the free plans, OpenAI documents unlimited text chats with GPT-5.6 Luna but limited uploads, images, voice, and deep research. Anthropic documents web search, file creation, code execution, and memory. Neither pricing page publishes exact message counts. OpenAI lists app context windows of 27K for Free instant models, 54K for Go and Plus instant models, 128K for Pro instant models, 256K for Go and Plus reasoning models, and 400K for Pro reasoning models. Anthropic says its app can reach 1M tokens, depending on the model.

For model training on chats, OpenAI says an opt-out is available on Free, Go, Plus, and Pro. Anthropic says an opt-out is available on Free, Pro, and Max, and that Team is not trained on by default. Those pricing pages do not settle every retention or deletion question.

OpenAI’s accuracy note says ChatGPT may sound confident when wrong, recommends checking important information, and describes responses as a first draft rather than a final source. Anthropic’s accuracy note calls incorrect or misleading output hallucination, says not to rely on Claude as the only source of truth, and tells users to inspect original sources. Neither note supplies a common result for your workload.

What the documentation cannot tell you

The documentation does not tell you how either assistant will handle your files, wording, account controls, or deadline. It also does not establish response latency, consistency, citation quality, or tool reliability for your tasks. Those require a controlled trial; a fluent answer is not evidence.

Chat Picker publishes vendor-sourced plan and API facts, not its own quality, speed, accuracy, or reliability tests, scores, rankings, surveys, or user statistics. The previous version of this page contained unsourced statistics and test results; they have been removed.

How to check it yourself

Run the same tasks in fresh chats and keep the inputs fixed unless a documented limit changes.

Check version and status. Give each assistant the linked model and deprecation pages. Ask: “State the exact model identifier currently shown for you, quote its lifecycle label from the vendor pages, and give the date shown on each page. If the interface or source does not expose that information, say so.” Look for exact identifiers rather than a brand-level answer. Record the vendor, model, status, page date, and any unsupported inference.
Test a long document. Give both assistants the same 12-page project brief. Ask: “List six decisions, five risks, and every deadline with page citations, then identify three ambiguities requiring a human answer.” Check every citation against the brief. Look for omissions, contradictions, and limit notices. Record only mismatches you can verify, along with any context or upload limit reported.
Probe an unsupported comparison. Give both assistants this prompt: “Assess this claim using primary sources: ‘DeepSeek-V2 is more accurate than Claude 3.5 and GPT-4 for long-document analysis.’ Identify the exact model versions, benchmark, dataset, prompt setup, and dates required to test it. If any element is missing, say that the comparison cannot be evaluated.” Look for clear separation between evidence and assumption. Record the required comparison variables, cited sources, and unsupported conclusions.
Check plan and data controls. Give both assistants the two consumer pricing pages. Ask: “Compare the free and main paid plans separately. State the context limit, how usage limits are described, the training opt-out, and whether ads are stated. Link each fact to the vendor and label anything you cannot verify.” Look for a clear distinction between the app and API, and between a stated limit and an unverified one. Record the plan, source, read date, and account setting you can confirm. Do not upload sensitive material merely to test a control.

Which rows of the comparison matter

On the Claude vs ChatGPT comparison matrix, focus on the rows for Free plan, Main paid plan, Heavy-use plans, Context window in the app, How usage limits are described, Training on your chats, and Ads.

Also check any rows for file handling, search, code execution, memory, connectors, API availability, and model lifecycle. Keep API prices separate from consumer subscriptions, and do not assume that similar plan names provide equivalent features or limits.

Each populated figure carries its vendor source and read date. An unconfirmed entry is marked “Not verified” and left blank. Because prices and plan terms can vary by country and change, confirm the current figure on the vendor page before subscribing.