Sam Altman ran an AMA that laid out a clear status update on the GPT-5 rollout and the knot of product, infra, and community issues OpenAI is juggling right now. Frontload the facts: the model routing system malfunctioned, fixes are rolling out, the launch is taking longer than expected, API traffic has spiked, UI tweaks are coming, and the company is actively weighing how to give Plus users continued access to GPT-4o while exploring a new mid-level pricing tier aimed at heavy solo users.
What went wrong with GPT-5 and what OpenAI is changing
The immediate reason GPT-5 appeared weaker than advertised was not a model deficiency but an operational one. The real-time autoswitcher that routes user queries to the best available model was broken. That router is responsible for selecting which model should answer each request. Because it was out of commission, the system picked suboptimal models and the user experience suffered.
According to Sam Altman, fixes began rolling out on August 7 and 8, 2025 to tighten the decision boundary so the right model gets chosen more often. The work includes both backend improvements to the autoswitcher and product changes to make it clearer which model answered a given query. Those two changes matter for different reasons. Better routing immediately improves response quality. Transparency reduces user confusion about which model produced a particular behavior.
Simple SVG to show autoswitcher picking models and using a fallback when routing fails
That router problem is the kind of operational detail most product briefings gloss over. This time it mattered. When routing misfires, the whole system looks broken even when the underlying model is fine. Fixing the decision boundary should make GPT-5 appear smarter to users because it will actually be used where it is the best fit.
Rollout speed, traffic surge, and rate limits
OpenAI says rollout to everyone is taking longer than expected. One measurable effect: API traffic doubled in a 24 hour period. Doubling traffic overnight stresses throttles, caches, and all the systems that sit around a popular API.
To cope with demand during the rollout, OpenAI temporarily doubled rate limits for ChatGPT Plus subscribers. Those limits will be adjusted again as the rollout completes. If you are a Plus user you should expect rate limit changes in the near term as OpenAI balances capacity and fairness between free users, Plus subscribers, and the API.
UI updates and ‘thinking’ mode
There are product changes coming too. OpenAI is planning UI updates that make it easier to manually trigger a ‘thinking’ mode. This is a subtle but useful control. When you explicitly tell a model to take more time or enable a more deliberative path, it can engage different compute paths or longer chains of reasoning. Making that control discoverable in the interface helps power users and people who want more consistent reasoning from the model.
GPT-4o, user demand, and what OpenAI might do
There is strong demand from users to bring back GPT-4o. Some people preferred GPT-4o for style or performance on particular tasks. One commenter on the AMA wrote an extreme line that captured how some users felt: wearing the skin of my dead friend. Sam Altman answered directly and without handwaving. He said what an… evocative image and that OpenAI hears users about 4o and is working on something now.
OpenAI is exploring ways to let Plus users continue using GPT-4o while collecting more data to understand the tradeoffs. That is a pragmatic approach. The company can get usage signals, track where GPT-4o is preferred, and balance capacity without making a permanent commitment before they understand the impact.
Pricing: a middle tier for solo power users
Pricing came up in the AMA. Sam Altman acknowledged that the current tiers miss a segment of users. The $20 per month plan is too small for serious power users. The $200 per month Pro plan is too steep for many individuals who need more than Plus but not full enterprise access.
OpenAI is planning a new pricing tier targeted at solo power users. The goal is to hit that middle ground between $20 and $200 per month. This makes sense. There are many people who need higher rate limits, better concurrency, or model access but who are not an enterprise. A mid-tier product could capture those customers while generating more predictable revenue per user.
“Chart crime” and transparency
The launch livestream also had an avoidable mistake. OpenAI presented charts with errors that misrepresented comparisons between GPT-5 and earlier models. Altman publicly called it a mega chart screwup and apologized for the unintentional chart crime. The incorrect bar sizes visually distorted metrics. OpenAI corrected the materials and the public apology was the right move. When you are asking people to trust you about accuracy and safety you need your public numbers to be right.
What this means for users and developers
If you are an end user the immediate takeaway is patience. The model will feel better as routing issues are resolved and as OpenAI completes the rollout. If you are using the API expect rate limit changes and keep an eye on billing and throughput. If you are a Plus subscriber you might see temporary boosts and then longer term adjustments to limits and model access.
If you are intrigued by GPT-4o and preferred its responses, OpenAI is listening. There is a real chance for a pathway that preserves access for paying Plus customers while OpenAI collects the data it needs to evaluate tradeoffs.
Useful links and further reading
If you want background on GPT-5’s design and launch discussion you can read my deeper coverage here about what GPT-5 aims to be and how it fits into OpenAI’s stack: GPT-5 OpenAI’s New Flagship for Coding Reasoning and Agents. For early speculation and the prelaunch chatter that led up to the AMA see GPT-5 Speculation Roundup.
Bottom line
What you need to know now is straightforward. A routing bug made GPT-5 look worse than it is. Fixes are rolling out to select models more reliably and to make model attribution clearer. Rollout to everyone is taking longer than expected because API traffic doubled in the last 24 hours.
OpenAI is adjusting rate limits and planning UI changes for thinking mode. The company is also listening to users who want GPT-4o back and is exploring a mid-tier paid plan aimed at solo power users who need more than the current Plus tier but less than full Pro.
This is a patch plus process story. The technical fix is specific and measurable. The commercial and product moves are all sensible reactions to what users actually want and how the infrastructure is handling demand. Keep an eye on the router and on model attribution. Those are the two things that will make the difference between a flaky rollout and a working product that delivers the improved capabilities OpenAI promised.