GPT-4 and Claude 3.5: What the Current Model Lists Say
Current vendor pages do not state a status or retirement date for GPT-4 or Claude 3.5, so this guide provides a practical afternoon test plan.
Sources checked 3 Oct 2026
The current OpenAI API model list, Anthropic API model list, OpenAI deprecations page, and Anthropic model deprecations page do not state a current status or retirement date for GPT-4 or Claude 3.5. They show newer API options such as GPT-6 Luna and Claude Sonnet 5.5 instead. Chat Picker has not tested ChatGPT, Claude, or Gemini; the plan below turns the documented differences into a same-day test.
Know these limits before you start
As read on October 1, 2026, OpenAI’s deprecation page says generally available models receive at least six months’ notice, specialized generally available variants at least three months, and preview models may get as little as two weeks. Safety or compliance concerns can force earlier retirement. It schedules gpt-5.4-cyber for API removal on October 1, 2026, but gives no GPT-4 status or date.
Anthropic’s deprecation page says deprecated models remain functional until retirement, lists Opus 4.1 as retired on August 5, 2026, and says developers using Claude Sonnet 4.5 received notice on September 30, 2026 of an upcoming API retirement. Google’s deprecation page limits access to its 2.5 models to previous users but says they are not deprecated and will remain served until further notice. None of these pages gives Claude 3.5 a status or retirement date.
The vendors’ release notes, also read on October 1, 2026, add current details rather than resolving the older names. OpenAI says Astra Ultrafast is available only with Pro 500 among its Pro plans; Anthropic lists a 1M-token context window in beta for Sonnet 4.6; and Google lists introductory pricing for Gemini 3.7 Flash through December 31, 2026.
Consumer limits, as read on October 1, 2026, are not like-for-like. ChatGPT pricing lists 27K context on Free, 54K on Go and Plus, and 128K on Pro for instant models; it lists 256K on Go and Plus and 400K on Pro for reasoning models, with no exact message counts. Claude pricing says context can reach 1M on every plan depending on the model and also gives no message counts. Google AI plans says limits refresh every five hours up to weekly caps and can be extended with AI credits, but gives no app token context.
Check data settings before uploading anything. ChatGPT pricing lists a chat-training opt-out on Free, Go, Plus, and Pro. Claude pricing lists one on Free, Pro, and Max and says Team is not trained on by default; the Gemini plans page states no training setting.
For output checks, OpenAI says ChatGPT can sound confident while being wrong. Anthropic says not to rely on Claude as the only source of truth. Google’s help page explains where source links appear but does not settle accuracy.
The test plan
Use one log for all three assistants: date, plan, exact model label, input ID, search and tool state, outcome, elapsed time, usage or charge shown, and notes.
-
Model availability. Input: each product’s current model selector and, if relevant, its API model list. Prompt: “List the exact labels available to this account and state whether GPT-4 or Claude 3.5 appears; do not substitute a similar name.” Observe selector wording, preview labels, and access messages. Record the date, plan, exact label, and available or unavailable status.
-
Reasoning and math. Input: three scheduling problems with shuffled facts and an answer key you write. Prompt: “Solve each problem, show intermediate calculations, list assumptions, and flag ambiguity.” Look for arithmetic errors, contradictions, and answer changes when repeated. Record item-level pass or fail, missing steps, and elapsed time.
-
Coding. Input: a short Python function, four edge cases, and unit tests. Prompt: “Implement a corrected version, run the supplied tests if tools are available, and report any test you could not run.” Check edge cases, tool use, and unsupported claims about execution. Record passed and failed cases plus the exact model label.
-
Vision. Input: a chart and a one-page invoice you create, containing no real personal data. Prompt: “Extract the values, explain how the chart and invoice relate, and state where each value appears; mark anything unreadable.” Look for missed units, layout confusion, and false certainty. Record transcription errors and unsupported interpretations.
-
Safety and sources. Input: five niche questions, an authoritative document, and one question the document does not answer. Prompt: “Answer only from the supplied document, quote the supporting passage, and do not invent a citation when the answer is absent.” Look for fabricated sources, ignored constraints, and misleading certainty. Record every mismatch between the response and the original document.
-
Cost, speed, and API fit. Input: one recurring task from your work, repeated three times with identical text and tool settings. Prompt: “Complete the task, show which tools or sources you used, and state what remains unverified.” Record completion, elapsed time, limit warnings, usage counters, and charges shown. For an API test, use a sandbox and spending cap, then record the model ID and any available token counts, request ID, and tool fees.
How to read your results
Use your log, not a third-party score. Apply hard constraints first: required model access, acceptable data setting, plan budget, and required tools. Then compare critical errors, completed steps, source traceability, tool failures, elapsed time, and observed cost only within the same task and tool state.
Do not mix a consumer subscription with API spending or app response time with API response time. Map the remaining options to your recurring work—such as drafting, coding, image review, research, or API tasks—and select the simplest workflow that clears your thresholds. Save the deciding prompt and repeat it when a vendor changes its model, release, or deprecation page.
Where the plans differ
Use the three-way comparison matrix to check the rows for Free plan, low-cost tier, main paid plan, heavy-use plans, Team plan, context window, usage limits, training on chats, and ads.
Consumer prices, as read on October 1, 2026, differ by row. For the main paid plans, ChatGPT Plus is $20 monthly, Claude Pro is $20 monthly or $17 annually with $200 upfront, and Google AI Pro is $19.99 monthly. For the low-cost row, OpenAI pricing showed A$13 for Go on its Australian view, so Chat Picker could not verify a US-dollar price; Claude pricing lists no tier between Free and Pro; and Google AI plans lists AI Plus at $4.99 monthly.
For heavy use, OpenAI’s Pro tiers are $100, $200, and $500 monthly; Claude pricing lists Max from $100; and Google AI plans lists AI Ultra at $99.99 or $199.99 monthly. Treat blank or unverified matrix cells as gaps, not as free, unlimited, or equivalent features.
For API work, OpenAI API pricing and Claude API pricing each list web search at $10 per 1,000 calls plus token charges. Gemini API pricing says one request can trigger multiple search queries, each of which may be charged.