OpenAI has officially rolled out GPT-5, the most advanced version of its language model yet — and the biggest surprise? It’s now available for free to all ChatGPT users.
Unveiled during a packed livestream by CEO Sam Altman, GPT-5 represents a major leap in AI capabilities, with improved reasoning, lower hallucination rates, better memory, real-world integrations, and deeper personalization.
RELATED: ChatGPT hits 700m weekly users, over 4x from last year
This launch moves ChatGPT closer to becoming not just a chatbot, but a real digital assistant that adapts to your style, handles tasks, and understands context like never before.
GPT-5: More than just faster or bigger
While GPT-4 was powerful, GPT-5 brings a whole new dimension: "agent-like" reasoning. It doesn’t just answer questions — it interprets, evaluates, and performs actions on your behalf.
Smarter reasoning and fewer mistakes
GPT-5 breaks down complex, layered prompts into manageable parts, processes them in parallel, and offers context-aware responses — often in a single try. It’s also more honest: when it doesn’t know, it tells you — reducing the risk of hallucinations or false confidence.
Hallucination rates (errors):
-
GPT-5 Thinking: 4.8%
-
GPT-4o: 20.6%
-
o3: 22%
It’s also much more proactive when discussing healthcare topics, flagging concerns or results without overstepping its bounds. On HealthBench, GPT-5 hallucinated just 1.6% of the time, a massive improvement.
New features powered by GPT-5
With GPT-5 now live in ChatGPT, users get access to several upgrades designed for both productivity and personality.
Vibe coding: Describe a feeling, get working code
A favorite among developers: GPT-5 excels at “vibe coding” — where you describe the mood or experience of a digital product, and it delivers fully functional HTML, CSS, and JavaScript.
This isn’t just about syntax. It interprets intent and design logic, making it possible to go from an idea to code in seconds.
It also tops performance charts:
-
SWE-bench Verified:
-
GPT-5: 74.9%
-
Claude Opus 4.1: 74.5%
-
Gemini 2.5 Pro: 59.6%
-
Multi-size access for all users
GPT-5 is now offered in three sizes:
-
gpt-5 (full model)
-
gpt-5-mini
-
gpt-5-nano (API only)
Free users start with the main GPT-5 model, and drop to mini only after hitting usage limits. Pro users ($200/mo) get unlimited access, plus exclusive access to GPT-5 Pro and GPT-5 Thinking, which deliver slower but deeper, tool-assisted answers.
ChatGPT gets real personalities and voices
Choose a chatbot that sounds like you
You can now give ChatGPT a tone and attitude that matches your preferences. Available “personalities” include:
-
Cynic: sarcastic, witty, dry humor
-
Listener: empathetic, echoing, gentle
-
Nerd: trivia-rich, curious, fact-focused
-
Robot: objective, emotionless, task-oriented
These flavors don’t compromise the quality of answers, but shape the tone — for example, productivity tips may sound snarky with Cynic, or softly encouraging with Listener.
Voice mode gets a serious upgrade
The new Voice Mode now features near-human fluidity and can adapt tone, pacing, and inflection based on your request — from calming and warm to fast and focused.
-
Plus subscribers get nearly unlimited voice responses
-
Free users get a few hours per day
-
The old “Standard Voice Mode” will be phased out in 30 days
And yes — custom GPTs now speak with your selected voice settings, too.
GPT-5 connects with your Google life
A huge productivity booster: ChatGPT Pro users can now link Gmail, Google Calendar, and Contacts. That means ChatGPT can:
-
Schedule meetings
-
Draft emails based on past threads
-
Check your availability automatically
-
Find and pull contact details
-
Suggest next steps from ongoing email convos
All without you having to prompt it explicitly.
⚠️ OpenAI promises minimal data access — only what’s necessary, only when it’s helpful.
GPT-5 vs competition: How does it stack up?
GPT-5 is now leading or competitive in most benchmark tests:
| Benchmark | GPT-5 Score | Competitor |
|---|---|---|
| GPQA Diamond (PhD-level science) | 89.4% | Claude Opus 4.1 (80.9%) |
| SWE-bench Verified (code) | 74.9% | Claude Opus 4.1 (74.5%) |
| Humanity’s Last Exam (tools mode) | 42% | xAI Grok 4 Heavy (44.4%) |
Although GPT-5 slightly underperformed on some agentic tasks (like airline navigation), it outperforms in key real-world coding and knowledge areas.
Safety and honesty at the core
OpenAI’s Safety Lead, Alex Beutel, shared that GPT-5 is significantly better at:
-
Detecting bad-faith prompts
-
Avoiding manipulative behavior
-
Staying transparent when uncertain
-
Balancing helpfulness with caution
This leads to fewer unsafe outputs, more relevant refusals, and a more trustworthy AI experience for users.
What free users get today
OpenAI is now giving free users access to:
✅ GPT-5 as the default model
✅ All core capabilities (writing, reasoning, coding)
✅ New personalities
✅ Limited Voice Mode access
✅ Auto model routing (ChatGPT picks the best model based on context)
That means everyday users get real value — not a stripped-down teaser.
Developers: Pricing and control
GPT-5 is also live in OpenAI’s developer API in 3 sizes:
| Model | Input Price | Output Price |
|---|---|---|
| gpt-5 | $1.25 / million tokens | $10 / million tokens |
Developers can also control response verbosity, letting them set how detailed GPT-5’s replies should be.
With GPT-5, OpenAI is bridging the gap between chatbots and intelligent agents. It writes better, thinks deeper, and adapts faster — whether you're coding, writing, scheduling, or just having a chat.
By offering GPT-5 to all users, OpenAI is also doubling down on its mission: making advanced AI accessible, useful, and safe for everyone — not just enterprise power users.
