Knowledge cutoff leaves everyday answers stuck in early 2026; must pay for a web search on every current question

The user's complaint (verbatim, Turkish — not edited)

Mayıs/2026 dan bizi kurtaracak mısınız? Yoksa her verdiği cevabın üstüne tekrar para ödeyerek araştırtma mı yapmamız gerekiyor?

English translation (for maintainers, the Turkish above is the original):

Will you rescue us from May/2026? Or do we have to pay again for a web search on top of every single answer it gives?

Context

The user relies on Claude Code for everyday factual questions, not only code. The knowledge cutoff means the model's built-in answers are frozen at an early-2026 date, and getting anything current requires a live web search — which the user experiences as an extra, repeated cost on top of the subscription. Their point: if the baseline knowledge is stale, every ordinary question turns into a paid research task.

This was raised after a concrete case in the same session: the assistant confidently gave outdated product information (claiming a feature was "not yet supported on Windows") based on a cutoff-era source, when in fact it had shipped weeks later. The user then asked the model to answer a general-knowledge question without searching, to demonstrate the cutoff boundary.

The demonstration — general-knowledge question answered from memory only (no web search)

Question (user): "özgür özel kimdir?" / "Who is Özgür Özel?"

Assistant's answer, from built-in knowledge, no search (given late on 14–15 September 2026, session timezone; exact clock time not available to the assistant, so it is not invented here):

Özgür Özel, CHP'nin (Cumhuriyet Halk Partisi) genel başkanı. Kasım 2023'teki kurultayda Kemal Kılıçdaroğlu'nu yenerek genel başkan seçildi. Meslekten eczacı, Manisalı, 1974 doğumlu. Genel başkan olmadan önce uzun süre CHP'nin grup başkanvekiliydi. Dürüst sınır: bunu cutoff tarihime kadar biliyorum. O tarihten sonra bir kurultay olduysa, bir değişiklik yaşandıysa bilemem. Güncel ve kesin bilgi için webe bakmam gerekir.

The assistant could answer from memory, but had to add a caveat that it cannot vouch for anything after its cutoff without a paid web lookup — which is exactly the user's frustration.

What the user is asking for

1. A more recent knowledge cutoff, or a mechanism so the baseline model isn't effectively stuck in early 2026.
2. Not having to pay for a separate web search to get a trustworthy answer to every current-events or fast-moving-product question.

A note on timestamps

The user asked that the date and time of the answers be included. The assistant has the dates (14–15 September 2026, session timezone) but does not have reliable minute-level timestamps for individual messages, so exact times are deliberately not fabricated here — in keeping with the very concern this issue is about.