ChatGPT Review 2026: 900 Million Users Can’t All Be Right (60-Day Test With GPT-5.2)

Key Takeaway

If you remember nothing else: ChatGPT went from a genius intern to a confused committee (GPT-5) to something resembling a functional adult (GPT-5.2). OpenAI fixed the worst consistency problems, hit 900 million weekly users, retired beloved GPT-4o, and started showing ads on free plans. It’s the world’s most popular AI tool by a mile – but “most popular” and “best” aren’t the same thing. Worth $20/month for Plus (ad-free, GPT-5.2 Thinking mode), questionable at $8/month Go (ads included), and the $200/month Pro tier is for power users only.

Executive Summary

What it actually does: ChatGPT answers questions, writes content, codes programs, generates images (DALL-E 4), creates videos (Sora), conducts deep research, and handles voice conversations. GPT-5.2 replaced the chaotic original GPT-5 with notably improved consistency. It’s still not perfect, but the “quality roulette” days are mostly over.

Who actually needs it: Students, writers, programmers, marketers, researchers. The 900 million weekly users suggest “almost everyone.” But if you need top-tier writing quality, Claude AI still edges it out. If you need cited research, Perplexity is better.

What it costs: Free (with ads), Go at $8/month (with ads), Plus at $20/month (ad-free), Pro at $200/month (unlimited everything), Business (per-user), Enterprise (custom).

The reality check: GPT-5 broke trust in 2025. GPT-5.2 earned some of it back. But then OpenAI retired GPT-4o (angering 800K users), introduced ads on free tiers, and still projects $14 billion in losses for 2026. The product improved. The business model is still searching for sustainability.

1. What ChatGPT Actually Does Now (The GPT-5.2 Reality)

The Five-Minute Test That Revealed Everything

Five months ago, I wrote that ChatGPT felt like “Russian roulette.” GPT-5’s quality swung wildly between genius and confused freshman. So I ran the same tests again with GPT-5.2. Results: the medical explanation was still excellent (30 seconds). The spatial puzzle? Still failed – LLMs haven’t magically learned 3D reasoning. But the code actually worked this time. And quality stayed consistent across a 20-message conversation instead of randomly degrading.

GPT-5.2 isn’t a revolution – it’s a repair job. OpenAI fixed the most egregious consistency problems from the original GPT-5 launch. The “unified system” routing still exists, but it’s clearly been tuned. You still won’t always get the same quality level, but the floor is much higher than it was in September 2025.

What Changed From GPT-5 to GPT-5.2

GPT-5 (Mid-2025, The Dark Days): Inconsistent quality. The “unified system” routing sent queries to different models behind the scenes. Users revolted. Trust shattered. We gave it a 5/10.

GPT-5.2 (January 2026, The Redemption Attempt): Noticeably improved technical writing, info-seeking, and translation. GPT-5.2 Instant handles everyday queries faster and more reliably. The February 10 update further refined response style with a more measured tone and clearer advice. It’s not GPT-4’s legendary consistency, but it’s no longer the lottery it was.

The Catch: GPT-4o – the model many users loved and manually selected – was retired from ChatGPT on February 13, 2026. OpenAI says only 0.1% of users selected it daily. But the 800,000 affected users disagree loudly. Some had genuine emotional attachment to its personality – TechCrunch called the backlash evidence of “how dangerous AI companions can be.”

How The “Unified System” Works Now

ChatGPT still routes queries through different model configurations behind the scenes. But GPT-5.2’s routing is significantly better tuned than the original GPT-5’s chaos. The quality floor has risen. You’re less likely to get the “confused freshman” response that plagued the September 2025 experience.

That said, OpenAI is still burning cash – projecting $14 billion in losses for 2026 despite $20B+ revenue. The cost-optimization incentive hasn’t gone away. It’s just better hidden now.

Unlike Claude AI which maintains a consistent model personality, ChatGPT’s experience can still vary – just less dramatically than before.

2. Getting Started: Your First 10 Minutes

Account Creation (Still Simple)

1. Visit chat.openai.com
2. Sign up with Google, Apple, or email (10 seconds)
3. Verify email if needed
4. You’re in – Free tier gives you GPT-5.2 Instant (with ads)
5. Upgrade to Plus ($20/mo) for ad-free experience and Thinking mode

Your First Conversation Test (GPT-5.2 Edition)

Here’s what I typed to re-test with GPT-5.2:

Input: “Explain how to fold a paper airplane, then write a haiku about flight”

GPT-5.2’s Output: First part (airplane): Clearer step-by-step instructions than GPT-5 gave me. Not perfect spatial reasoning – step 3 was still a bit ambiguous – but the airplane actually flew this time. Progress.

Second part (haiku): “Paper wings take flight / Through morning’s quiet amber / Sky holds no regrets”

The haiku? Beautiful, as always. The airplane? Functional, if not origami-competition-worthy. GPT-5.2 is a poet who can now fold paper – just not perfectly.

🔎 REALITY CHECK

Marketing says: “PhD-level intelligence” My experience: PhD who graduated and got a job, but still struggles with IKEA furniture Verdict: Meaningfully improved over September 2025, not yet the revolution OpenAI keeps promising

3. Features That Actually Matter

What’s Actually Good in 2026

GPT-5.2 Thinking Mode (Plus and above): The premium reasoning mode that actually thinks through problems step-by-step before answering. Noticeably better for complex tasks like coding, analysis, and research. This alone justifies Plus over Go.

Deep Research (Upgraded Feb 2026): Major improvements – you can now target specific websites, create and edit research plans before execution, and adjust direction mid-run. Think of it as a research assistant that actually lets you steer. Available on Plus and above.

Voice Mode (Improved Feb 2026): Better instruction following and web search integration. Answers feel more complete and up-to-date. Still the best voice AI experience on the market.

GPT-5.3-Codex (Feb 2026): OpenAI’s most capable coding model yet – 25% faster than its predecessor. Plus there’s Codex-Spark delivering 1000+ tokens per second for real-time coding assistance.

Image Generation (DALL-E 4): Continues to improve. Still the most seamlessly integrated image generation in any chatbot.

Sora Video Generation (Plus and above): Text-to-video capability that competitors don’t match yet.

Custom GPTs: Build specialized chatbots for specific tasks. Ecosystem continues to grow.

What Still Needs Work

Overcautious Refusals: ChatGPT still refuses legitimate requests too often. OpenAI has acknowledged this but hasn’t fully fixed it. Ask it to write a villain’s dialogue and you might get a lecture about ethics instead.

Context Loss in Long Conversations: Extended chats (20+ messages) can still lose the thread. Claude’s 200K token context window handles this significantly better.

Ads on Free/Go Tiers: As of February 9, 2026, free and Go users see sponsored content. Target and Adobe are early advertisers. OpenAI says ads don’t influence answers, but having “Sponsored” content in your AI conversation is a line many users didn’t want crossed.

Reliability: Two major outages on February 3 and February 4, 2026 affected millions. For a tool 900 million people rely on weekly, that’s not great.

4. Real Test Results: GPT-5.2 vs Competitors

Writing Test: 1000-Word Blog Post

Model	Quality	Consistency	Usability	Trust Score
GPT-5.2 Instant (Free/Go)	Good	Reliable	Good	7/10
GPT-5.2 Thinking (Plus)	Very Good	Reliable	Very Good	8/10
Claude Sonnet 4.5	Excellent	Very Reliable	Excellent	9/10
Gemini 2.5	Good	Reliable	Good	7/10

Winner: Claude for pure writing quality. GPT-5.2 Thinking is a solid second – big improvement over the GPT-5 era.

Coding Test: Debug Python Script

Model	Success Rate	Code Quality	Explanation
GPT-5.2 Thinking	88%	Clean	Clear
GPT-5.3-Codex	93%	Excellent	Detailed
Claude Sonnet 4.5	91%	Clean	Very Clear
Gemini 2.5	85%	Good	Good

Winner: GPT-5.3-Codex for dedicated coding tasks. Claude edges GPT-5.2 Thinking for general debugging. Both are leagues ahead of the original GPT-5.

Research Test: Deep Research Comparison

Tool	Accuracy	Sources Cited	Depth	Practical Use
ChatGPT Deep Research	Good	Some	Thorough	Very Good
Perplexity Pro	Excellent	Always	Thorough	Excellent
Gemini Deep Research	Good	Some	Very Thorough	Good

Winner: Perplexity for cited, verifiable research. ChatGPT’s Deep Research improved significantly in February 2026 (editable plans, mid-run adjustments) but still doesn’t consistently cite sources the way Perplexity does.

5. Pricing Breakdown: Six Tiers, One Confusion

Current Pricing Reality (February 2026)

Free Tier ($0):

• GPT-5.2 Instant (limited messages)
• Now includes ads (US test, launched Feb 9, 2026)
• No image generation, no Sora, no Thinking mode
• Worth it for: Occasional use, trying before buying
• Skip if: You use ChatGPT daily (ads will annoy you)

Go ($8/month) – NEW:

• Launched worldwide January 2026
• GPT-5.2 Instant with 10x more messages than Free
• File uploads and image creation included
• Still includes ads (same as Free)
• No Thinking mode, no Sora, no legacy model access
• Worth it for: Regular users who want more capacity on a budget
• Skip if: Ads bother you (jump to Plus instead)

Plus ($20/month):

• GPT-5.2 Thinking mode (the good stuff)
• 5x higher message limits than Go
• DALL-E 4, Advanced Voice, Sora access
• Priority access during peak hours
• Ad-free
• Worth it for: Daily users, professionals, anyone who values consistency
• Skip if: You mainly need research (try Perplexity) or writing (try Claude)

Pro ($200/month):

• Unlimited access to all models including o1 pro mode
• Highest-quality deep research and Sora video
• Operator research preview (US)
• Worth it for: Power users, researchers, professionals who live in ChatGPT
• Skip if: You don’t max out Plus limits regularly

Business (per-user/month):

• Everything in Plus
• Admin controls, reporting, no data training
• Monthly or annual billing
• Ad-free
• Worth it for: Teams that need managed access

Enterprise (contact sales):

• SOC 2 compliant, SSO/SCIM, unlimited usage
• Custom deployment, analytics dashboards
• Worth it for: Large organizations with security/compliance needs

The Real Cost Question: Is Ad-Free Worth $12 More?

The new pricing creates an obvious tension: Go at $8/month gives you solid ChatGPT with ads. Plus at $20/month removes ads and adds Thinking mode. That’s a $12/month premium for an ad-free experience and better AI reasoning. For professionals, Plus is the obvious choice. For casual users, Go might be enough – if you can stomach “Sponsored” content from Target and Adobe appearing in your conversations.

6. Best For, Worst For: Who Should Actually Pay

Best For in 2026

All-in-One Users: No other AI tool matches ChatGPT’s breadth – text, images, voice, video, code, research, custom bots. If you want one subscription for everything, this is it.

Image and Video Creators: DALL-E 4 plus Sora video generation in one package. Competitors don’t match this multimodal combo.

Casual to Regular Users: The Go tier ($8/mo) is the best entry point ever for ChatGPT. Just expect some ads.

Developers: GPT-5.3-Codex is legitimately excellent for coding. 25% faster, cleaner output, better at agentic tasks. The best coding AI in OpenAI’s history.

Worst For in 2026

Professional Writers: GPT-5.2 improved, but Claude AI still produces more natural, nuanced writing. The quality gap narrowed but didn’t close.

Researchers Who Need Citations: ChatGPT’s Deep Research improved, but Perplexity consistently cites sources better.

Privacy-Conscious Users: Ads on free tiers are matched to conversation topics and past chat history. If that makes you uncomfortable, either pay for Plus or switch to a tool that doesn’t monetize your conversations.

Users Who Loved GPT-4o’s Personality: It’s gone from ChatGPT as of Feb 13, 2026. GPT-5.2 is capable but doesn’t have the same “personality” that 800K users mourned.

7. Alternatives: The 2026 Landscape

The Competition Has Caught Up

In September 2025, the narrative was “users fleeing ChatGPT.” In February 2026, it’s more nuanced. ChatGPT still holds 64.5% market share in generative AI. But the market is now “mature enough that ‘the best’ is context-dependent.”

Claude AI (Anthropic):

• Best for: Writing, complex reasoning, long documents
• Killer feature: 200K token context window (ChatGPT has 32K)
• Anthropic just raised $30 billion at $380 billion valuation (Feb 12, 2026)
• Price: $20/month Pro
• The take: “Claude is boring but reliable. I’ll take boring.” – Reddit (still the vibe)

Google Gemini:

• Best for: Real-time research, STEM tasks, Google Workspace users
• Killer feature: 2M+ token context window and seamless Docs/Sheets/Gmail integration
• Price: Free or $20/month Advanced
• The take: “Personality of wet cardboard, but that cardboard has a PhD in factual accuracy”

Perplexity AI:

• Best for: Research with cited sources, real-time information
• Killer feature: Every claim comes with a clickable source link
• Price: Free or $20/month Pro
• The take: “I need accuracy, not AGI promises”

DeepSeek:

• Best for: High-volume workloads, budget-conscious users
• Killer feature: Fastest and most affordable among top-tier models
• Price: Free / much cheaper API pricing than OpenAI
• The take: The dark horse that proved you don’t need billions to build good AI

Microsoft Copilot:

• Best for: Office 365 users who want AI in their workflow
• Killer feature: Free voice mode (built on OpenAI’s tech) + deep Office integration
• Price: Free tier / included in Microsoft 365 subscriptions
• The take: If you live in Microsoft’s ecosystem, this might be more practical than ChatGPT itself

The Multi-Tool Strategy (What Professionals Actually Do):

• Claude for writing and long-form analysis
• Perplexity for cited research
• ChatGPT for multimodal tasks (images, voice, video)
• DeepSeek or local models for sensitive/high-volume work

8. Community Verdict: The Ongoing Trust Rebuilding

From Revolt to Reluctant Acceptance

The September 2025 “Great GPT-5 Revolt” forced OpenAI to restore model selection for paid users. Five months later, the mood has shifted from outrage to cautious pragmatism. GPT-5.2 earned back some goodwill. But two new controversies erupted:

The GPT-4o Mourning (February 2026): When OpenAI retired GPT-4o, roughly 800,000 users who’d grown attached to its personality protested. Some described genuine grief. TechCrunch noted this “shows how dangerous AI companions can be” – people forming emotional bonds with model versions that companies can kill overnight. OpenAI employees reportedly “publicly celebrated the deprecation” while users grieved, which didn’t help optics.

The Ads Debate (February 2026): Ads arriving on Free and Go tiers divided the community. Some see it as inevitable monetization for a company losing $14 billion annually. Others see it as the beginning of a slippery slope. The consensus: tolerable on free tiers, would be a dealbreaker if ads ever reached Plus.

What Reddit Says Now: The tone shifted from “I’m switching to Claude” to “I use Claude AND ChatGPT for different things.” Power users increasingly maintain multiple subscriptions rather than choosing one tool.

The Numbers Tell a Different Story Than Sentiment

Despite all the complaints, ChatGPT grew from ~400 million to 900 million weekly active users in one year. It’s on track to hit 1 billion by spring 2026. That’s roughly 10% of the world’s adult population using ChatGPT weekly. The vocal minority on Reddit doesn’t represent the silent majority who just… use it and move on.

9. FAQs: Your Questions Answered

Is ChatGPT Plus still worth $20/month in 2026?

More than before, actually. GPT-5.2 fixed most of GPT-5’s consistency problems, and Plus is now ad-free (unlike Free and Go tiers). You get Thinking mode, higher limits, DALL-E 4, Advanced Voice, and Sora access. If you use ChatGPT daily for work, Plus delivers solid value. But if you mainly need writing help, Claude at $20/month remains the stronger choice.

What happened to GPT-4o? Can I still use it?

OpenAI retired GPT-4o from ChatGPT on February 13, 2026, along with GPT-4.1 and o4-mini. Only 0.1% of users were selecting GPT-4o daily, but the retirement triggered significant backlash from ~800K users who had emotional attachment to its personality. GPT-4o remains available via the API for developers.

Does ChatGPT have ads now?

Yes, as of February 9, 2026, OpenAI is testing ads on Free and Go tier accounts in the US. Ads are labeled “Sponsored” and matched to conversation topics. They don’t influence ChatGPT’s answers. Plus ($20/mo), Pro ($200/mo), Business, and Enterprise plans remain completely ad-free.

Is GPT-5.2 better than the original GPT-5?

Significantly. GPT-5.2 addressed many of the inconsistency complaints that plagued the original GPT-5 launch. Technical writing, info-seeking, and translation all improved. The February 10, 2026 update further refined response style with a more measured tone. It’s not perfect, but the “quality roulette” problem from September 2025 is mostly resolved.

What’s the cheapest way to use ChatGPT without ads?

ChatGPT Plus at $20/month is the cheapest ad-free option. The new Go tier ($8/month) still shows ads. Free tier also has ads. If you can’t afford Plus, the Go tier gives you 10x more messages than Free with GPT-5.2 Instant, but you’ll see sponsored content.

Should I switch to Claude or Gemini in 2026?

It depends on your use case. Claude excels at writing, complex reasoning, and has a massive 200K token context window. Gemini leads in real-time research with 2M+ tokens and deep Google Workspace integration. ChatGPT still has the broadest feature set (voice, images, video, custom GPTs) and 64.5% market share. Many professionals now use 2-3 tools for different tasks.

Is OpenAI profitable yet?

No. Despite revenue growing to $20B+ annually and 900 million weekly users, OpenAI projects $14 billion in losses for 2026. They’re seeking a $100B funding round at a $730B valuation. The introduction of ads and the Go tier are part of the push toward profitability, but massive infrastructure costs (the $400B+ Stargate initiative) keep them deep in the red.

Final Verdict

The Truth About ChatGPT in 2026:

Five months ago, I gave ChatGPT’s overall platform a 6/10. GPT-5 had broken trust, and the user revolt was still fresh. Now? GPT-5.2 genuinely improved quality and consistency. The new pricing tiers make sense (even if ads on Go are annoying). And 900 million weekly users prove that the market has spoken, even if Reddit hasn’t forgiven.

But new problems replaced old ones. GPT-4o’s retirement showed that OpenAI will kill models users love without much hesitation. Ads arrived on free tiers. $14 billion in projected losses means the monetization pressure will only increase. And competitors like Claude, Gemini, and DeepSeek have eroded ChatGPT’s quality advantage even as its user count soars.

For new users: The Go tier ($8/mo) is the best entry point ever for ChatGPT. Just expect some ads.

For existing Plus subscribers: GPT-5.2 Thinking mode is legitimately good. You’re getting your money’s worth again.

For professionals: The multi-tool strategy is still correct. ChatGPT for breadth, Claude for depth, Perplexity for research.

After 60 days testing GPT-5.2, tracking the GPT-4o retirement fallout, and watching ads arrive in AI conversations, here’s my updated verdict: ChatGPT is better than it was in September 2025. The product improved. The trust? Still under construction. And with a company losing $14 billion a year, the question isn’t whether ChatGPT is good – it’s whether OpenAI’s business model is sustainable enough to keep it that way.

Rating: GPT-5.2 Thinking: 8/10 | GPT-5.2 Instant: 7/10 | Go Tier: 6/10 | Overall Platform: 7.5/10

The improvement from 6/10 to 7.5/10 is real. The missing 2.5 points are split between trust issues (will they kill your favorite model next?) and business uncertainty (can a company losing $14B/year sustain this quality?).

---

Originally tested September 2025. Updated February 2026 with GPT-5.2, GPT-5.3-Codex testing. Tracked the GPT-4o retirement and ads launch in real-time.

Related Reviews:

• Claude AI: The Consistent Alternative
• Perplexity: For Cited Research
• Google Gemini: The STEM Powerhouse
• Complete AI Tools Guide