OpenAI’s GPT-5 Rollout – What It Means for SMBs, Entrepreneurs, and Everyday Users

OpenAI’s GPT-5 is here, and I sat down to explore what it could mean for small businesses, entrepreneurs, and general users.

First impressions: The reports, reviews, and early experiences are mixed. Separating emotion from fact is tricky at this stage, and more testing is needed. I’ll update this post as I learn more.

Here’s what I’ve gathered so far:

Hype
Sam Altman likens the jump between models to moving from a high school student (GPT-3) to a college student (GPT-4) to a PhD-level expert (GPT-5).

Model availability

• Free users can no longer select older models.

• Paid subscribers can choose between GPT-5 and GPT-5 Thinking.

• To make GPT-4o available in the model selector, go to Settings > General and toggle Show legacy models.

Auto-routing
GPT-5 uses an auto-routing system that analyzes query complexity and context to select the best internal sub-model. This makes it faster and more context-aware, delivering quick answers or deep reasoning depending on the request.

Context window sizes

• Free: 8k tokens

• Plus: 32k tokens

• Pro: 128k tokens

• API: Up to 400k tokens (272k input, 128k output)

Code generation
Improved at writing and editing code, especially when paired with OpenAI’s Canvas app for live previews. (See comparison with Claude Opus 4.1.)

Multimodality
Better at interpreting handwriting and visuals, now supporting full multimodal input: text, voice, video, and images. It can understand screenshots and short video clips.

Healthcare focus
OpenAI has prioritized reliability and accuracy in medical queries, making GPT-5 a stronger informational partner (but still not a physician substitute).

Personalities
Customizable text-mode personalities Cynic, Robot, Listener and Nerd are now available. These don’t affect Voice Mode.

Memory
ChatGPT 5 remembers user preferences, previous conversations, and context for longer, reducing the need for repeated reminders or re-explaining tasks.

Agents
GPT-5 can automate multi-step tasks, acting as an AI “autopilot” for things like scheduling or project management.

Usage limits
Free tier:

• 10 GPT-5 messages every 5 hours (48/day, 336/week)

• 1 GPT-5 Thinking message/day (7/week)

Plus tier:

• 80 GPT-5 messages every 3 hours

• 200 GPT-5 Thinking messages/week

Pro tier:

• Unlimited GPT-5 and GPT-5 Thinking usage

Connectors
Pro users can now link Gmail, Google Contacts, and Google Calendar to ChatGPT - available beyond Agent mode. ChatGPT automatically uses these when relevant, so you don’t have to manually select them. Lower tiers will gain access later.

Hallucinations
GPT-5 is less prone to fabricating answers and better at stating when it lacks sufficient information. False success claims (“deceptions”) have also been reduced.

Early observations

• Some users report inconsistencies. For example, the model allegedly failed the “how many b’s in ‘blueberry’?” test. Tt worked fine for me, but it automatically switched to GPT-5 Thinking.

• Deepak Paramanand notes in a comment on LinkedIn (echoing Andrej Karpathy) that ‘LLMs often fail at letter-level tasks due to tokenization methods that don’t maintain single-letter precision’.

• New rate limits have been described as “generous,” though not without controversy when reading between the lines.

• GPT-5 was trained in part with synthetic data generated by o3 which bears interesting implications regarding model degradation and bias amplification.

Overall, it appears that GPT-5 is a quality-of-life update that brings some welcome improvements to performance, safety and usability. Enhanced coding and writing skills, persistent memory, multimodal input, specialized personalities, and safer and clearer responses to ambiguous queries generally mark improvements while the new automatic model switching should also fall into that category. One thing is clear: Effective and more nuanced prompting is more important than ever to get the best results out of this model.