For the past year or so, being a free ChatGPT user meant watching the paid crowd enjoy the good stuff while you made do with whatever OpenAI left on the table. Today, that gap got a little smaller — and OpenAI’s plans for the next six months suggest it’s about to get even more interesting.
Meet GPT-5.4 Mini and Nano
OpenAI just released GPT-5.4 mini and nano — two smaller, faster, cheaper versions of the GPT-5.4 model that launched a couple of weeks ago for higher-tier subscribers. And this time, free users actually get a piece of the action.
GPT-5.4 mini is now available to Free and Go ChatGPT users via the “Thinking” feature in the + menu. That’s not just a marketing gesture — this is a genuinely better model. It runs more than 2x faster than the previous GPT-5 mini, and in several benchmarks it gets surprisingly close to the performance of the full GPT-5.4 model.
Here’s how the numbers shake out:
- SWE-Bench Pro (coding): GPT-5.4 mini scores 54.4% vs. 57.7% for the full GPT-5.4 — and 45.7% for the old GPT-5 mini. That’s a meaningful jump.
- OSWorld-Verified (computer use): 72.1% for mini vs. 75.0% for the full model. Basically within spitting distance.
- GPQA Diamond (hard science reasoning): 88.0% vs. 93.0%. Still very solid.
The nano model is the even tinier version — aimed at classification, data extraction, ranking, and cheap subagent tasks. It’s not trying to write your thesis; it’s trying to sort your inbox at roughly the cost of a postage stamp per million tokens.
Why This Actually Matters
OpenAI has been pretty explicit about what these models are designed for: speed-sensitive workflows where you can’t wait around for a big model to think. Coding assistants that need to feel snappy. Multimodal apps that interpret screenshots in real time. Subagent systems where a large model plans and smaller models execute.
That last use case is worth pausing on. The new “subagent” architecture means a big model like GPT-5.4 can handle the strategy — deciding what to do, coordinating the pieces — while delegating grunt work to GPT-5.4 mini running in parallel. It’s less “one AI doing everything” and more “a team of AIs with different job descriptions.”
In Codex (OpenAI’s coding agent), GPT-5.4 mini only uses 30% of the GPT-5.4 quota, so power users get to run more tasks without burning through their limits. In the API, mini costs $0.75 per million input tokens and $4.50 per million output tokens — with a 400k context window, which is substantial for the price.
As 9to5Google noted, the timing also overlaps with the rise of “vibe coding” — the trend of building software by describing what you want rather than writing every line manually. GPT-5.4 mini handles “targeted edits, codebase navigation, front-end generation, and debugging loops with low latency.” Translation: it can keep up with you when you’re in a flow state and don’t want to wait three seconds for every autocomplete.
But Wait, There’s a Bigger Story
The mini and nano launch is the present-tense news. The future-tense news is arguably more interesting.
According to recent reports, OpenAI plans to deploy an autonomous AI research intern by September 2026. Not a tool that helps researchers. An intern. One that can take on research tasks semi-independently, the kind of thing you’d hand off to a junior hire and check in on later.
And that’s apparently just the intermediate milestone. OpenAI’s stated “North Star” is building a fully automated multi-agent research system by 2028 — a setup where AI agents run the research pipeline end to end, from literature review to experiment design to write-up, without a human needing to babysit every step.
That’s a significant ambition to state publicly. It’s the kind of goal that either sounds visionary or delusional depending on how you feel about how AI has been progressing lately. Given the trajectory of these models — each release genuinely faster and more capable than the last — maybe it’s not as far-fetched as it sounds.
What Free Users Should Actually Do Right Now
If you’re on the free tier and haven’t tried the Thinking feature yet, now’s a decent moment to poke at it. GPT-5.4 mini brings real reasoning improvements, not just a new version number. The difference is most obvious on technical tasks — coding, math, anything where you need the model to work through steps rather than just pattern-match to an answer.
For developers, the API pricing makes mini genuinely worth considering as a workhorse model for high-volume tasks. The 400k context window means you can throw substantial documents at it without chunking nightmares. And at $0.75/M input, you can afford to be less precious about prompt length.
The Bottom Line
OpenAI is doing two things simultaneously: widening access to better AI for regular users (GPT-5.4 mini in the free tier) and laying the groundwork for AI that does increasingly independent cognitive work (the research intern roadmap). These aren’t unrelated. The same architectural improvements that make mini fast and capable are what make the subagent stack work — and what, eventually, might power something that does research while you sleep.
Whether the September deadline holds is anyone’s guess. But the direction is clear, and the pace isn’t slowing down.
Sources: OpenAI, 9to5Google
🐾 Visit the Pudgy Cat Shop for prints and cat-approved goodies, or find our illustrated books on Amazon.