GPT-4.1 Has a 1M Token Context–ChatGPT Still Limits Users to 32K

1 Asmordikai 0 5/27/2025, 6:17:23 AM

I’m a long-time ChatGPT Plus subscriber and started using GPT-4.1 in ChatGPT as soon as it was released.

One of GPT-4.1’s core features is its 1 million token context window. But in ChatGPT, GPT-4.1 is still limited to 32,000 tokens—the same as GPT-4o.

This limitation isn’t clearly disclosed in the ChatGPT UI, subscription page, or rollout announcement. You only learn it’s API-only by reading developer documentation or forum posts.

If the model supports 1 million tokens and OpenAI already delivers that through the API, why is the core user-facing product—ChatGPT—still capped at 32K?

At a minimum, OpenAI should:

Clearly state token limits within the ChatGPT interface.

Stop advertising 1M-token context in GPT-4.1 materials without a disclaimer.

Provide a roadmap to enable full context support in ChatGPT.

Would appreciate hearing how others feel about this. Shouldn’t Plus users have access to what they’re paying for?

Guide to Fiber Optic Network Design [pdf] (thefoa.org)

The Blowtorch Theory: A New Model for Structure Formation in the Universe (theeggandtherock.com)

Show HN: I made an exclusive list of Ahrefs alternatives (AhrefsAlternative.com) (ahrefsalternative.com)

Decoding TCP SYN for Stronger Network Security (netscout.com)

Meme retailer GameStop buys first Bitcoin batch, scooping up $500M (cnbc.com)

Rust to WebAssembly the Hard Way (surma.dev)

Sodium-air fuel cell for high energy density and low-cost electric power (cell.com)

The Largest Gathering in Human History (googlemapsmania.blogspot.com)

Cory Doctorow on how we lost the internet (lwn.net)

AI generates Architectural Blueprint from scratch (youtube.com)

Building Search Index with Eleventy (arielsalminen.com)

AtomVM, the Erlang virtual machine for IoT devices (atomvm.net)

Ask HN: What do you love about your job?

Work When Your Team Works: How to Align Time Zones in On-Demand Talent (theflock.com)

Turn any PDF into a published, editable docs site (docshound.com)

Using SAT to Get the World Record on LinkedIn's Queens (ryanberger.me)

Texas' annual reading test adjusted difficulty yearly, masking improvement (theconversation.com)

Show HN: Icepi Zero – The FPGA Raspberry Pi Zero Equivalent (github.com)

Qt's New Bridging Technology – Looking Back to Move Forward (qt.io)

The Useless Web – click a button to visit a random "useless" site (theuselessweb.com)

Microcombs could be key in meeting bandwidth demands for AI and data centers (techxplore.com)

Please proof with caution: a rebuttal to a "proof" against free will (souldzin.substack.com)

Don't solve problems you don't have. You're creating problems (ohhfishal.net)

SteamOS Destroys Windows (pointieststick.com)

Show HN: Manage forms as code – dead simple syntax, AI-friendly (form-data.com)

The Electric Telegraph, an Expeditious Method of Conveying Intelligence (2016) (technicshistory.com)

Show HN: A self-reflective system that rewrites itself as it thinks (reflexivetotality.com)

Show HN: SCANOSS Now Scans Container Images

Ask HN: Do tech companies avoid hiring cynical candidates?

The ESPN Streaming Service, the Status of U.S. Sports Rights, the Danger Of (stratechery.com)

Ways to Optimize Your Code in Python (blog.appsignal.com)

New Linux Patches Properly Handle Audio Jack on Sony PS5 DualSense Controller (phoronix.com)

Trafilatura: A tool and library to gather text and metadata on the Web (github.com)

Yambda-5B – a large-scale multi-modal dataset for ranking and retrieval (huggingface.co)

Royal Jelly (en.wikipedia.org)

How Huawei built a 5nm chip under sanctions (twitter.com)

The Who Cares Era (dansinker.com)

Show HN: Ts-migrating – Progressively Upgrade tsconfig.json (github.com)

Combination therapy with Rapamycin and Trametinib prolongs the life of mice (mpg.de)

WhatsApp Launches Official iPad App (9to5mac.com)

Even Faster String Matching in Python (bergvca.github.io)

Brain drugs can now cross the once impenetrable blood–brain barrier (nature.com)

Show HN: Puck – Open-source visual editor for React (github.com)

Ask HN: Will countries ban realtime language translation to reduce overt racism?

Scrambling Up the Supply Chain (blog.supermechanical.com)

Cubic: Code Review for the AI Era (cubic.dev)

I built workflowhub AI for helping creators to monetize their automations

Custom domains and SSL in Rails development (avohq.io)

OpenSIPS Summit 2025 [video] (youtube.com)

The Buyer's Guide to Software Defined Object Storage (blog.min.io)

GPT-4.1 Has a 1M Token Context–ChatGPT Still Limits Users to 32K

Comments (0)