I spent way too much time this year switching between AI tools. ChatGPT for one task, Claude for another, Gemini because it was right there in my Gmail. By December, I had subscriptions to three different services and still wasn’t sure which one was actually the best.
Sound familiar?
2025 was the year the AI race got real. OpenAI released three major versions of GPT-5. Anthropic’s Claude went from niche developer favorite to mainstream contender. Google finally figured out how to make Gemini useful by putting it everywhere. And Perplexity quietly built something that might change how we search forever.
So who actually won 2025? I dug through the releases, the benchmarks, the controversies, and the real-world performance to figure it out.
Here’s my honest ranking.
The Contenders: A Quick Overview of the 2025 AI Race
OpenAI shipped GPT-5 through GPT-5.2 in five months, Anthropic grew revenue from $1 billion to $5 billion behind Claude 4 and Claude Code, Google embedded Gemini across Gmail and Android, and Perplexity’s valuation jumped from $520 million to $18 billion on 780 million monthly queries.
Before we get into winners and losers, let’s recap what each company actually shipped this year.
OpenAI (ChatGPT)
OpenAI came out swinging with GPT-5 in August, GPT-5.1 in November, and GPT-5.2 in December. That’s three major model releases in five months. They also launched ChatGPT Agent (an AI that can actually do things on websites), the Atlas browser, Sora 2 for video generation, and GPT-5.2-Codex for developers.
On paper, impressive. In reality, it was messier than the press releases suggested.
Anthropic (Claude)
Anthropic released Claude 3.7 Sonnet in February, then dropped Claude 4 (Opus and Sonnet) in May. By year’s end, they’d shipped Opus 4.1, Sonnet 4.5, Haiku 4.5, and Opus 4.5. Their revenue grew from $1 billion to over $5 billion in eight months. Claude Code, their coding assistant, hit $1 billion in revenue on its own.
They also created MCP (Model Context Protocol) and donated it to the Linux Foundation. More on why that matters later.
Google (Gemini)
Google released Gemini 2.0 Flash in February, Gemini 2.5 in March, and Gemini 3 Flash in December. They won a gold medal at the International Mathematical Olympiad. But more importantly, they put Gemini everywhere. Gmail, Google Photos, Chrome, Android, Search, Google Docs, and basically every product with a Google logo.
They also announced that Google Assistant will be replaced by Gemini in 2026. The integration play was aggressive.
Perplexity
The scrappy underdog launched Comet (their own browser) in October, released the Sonar AI model that’s 8x faster than their previous engine, and open-sourced R1-1776 under an Apache 2.0 license. Their valuation jumped from $520 million to $18 billion. Monthly queries tripled to 780 million.
On the Humanity’s Last Exam benchmark, Perplexity’s Deep Research scored 21.1%, beating OpenAI’s o1 (18%) and Google’s Gemini (19%).
4th Place: OpenAI (Yes, Really)
OpenAI placed last in this ranking because Sam Altman admitted the GPT-5 launch was “botched,” an internal “Code Red” memo revealed panic about Anthropic and Google, and the rapid-fire model releases felt like “ship it and fix it later” rather than careful product development.
This might be controversial, but hear me out.
OpenAI had a rough 2025. Sam Altman admitted the GPT-5 launch was “botched” after users complained the new model felt colder and less helpful. They had to bring back GPT-4o as an option because the backlash was so strong.
Internally, Altman sent a “Code Red” memo about increased competition from Google and Anthropic. CNBC reported that OpenAI was pulling back investments in health, shopping, and advertising to focus on improving ChatGPT’s core experience.
There were also the safety concerns. A lawsuit from parents who alleged ChatGPT encouraged their son’s suicide led to new protocols for users under 18. Altman himself admitted that “the potential impact of models on mental health was something we saw a preview of in 2025.”
My honest take. OpenAI is riding on brand recognition the way eBay did in the early 2000s. They were first. Everyone knows the name. But being first doesn’t mean you’re still the best.
When I use ChatGPT now, it feels like they’re trying to do everything (video, images, agents, voice, browsing) without doing any one thing exceptionally well. The experience is fragmented. The pricing is confusing (Plus? Pro? Enterprise?). And the rapid-fire model releases feel more like “ship it and fix it later” than careful product development.
GPT-5.2 is genuinely impressive on benchmarks. But benchmarks aren’t everything. In daily use, ChatGPT often feels like it’s lost the plot.
3rd Place: Perplexity
Perplexity’s valuation grew 34x in under two years with search accuracy at 91.3% and a hallucination rate below 3.5%, but their single-product focus on search limits them as a platform compared to more versatile competitors.
Perplexity had an incredible year for a company most people still haven’t heard of.
Their valuation grew 34x in under two years. Their search accuracy improved to 91.3%. Their hallucination rate dropped below 3.5%, which is genuinely impressive for an AI search engine.
The Comet browser launch was smart. Instead of asking people to change how they search, they built a browser that makes AI search the default experience. Millions were on the waitlist before it went free in October.
Where Perplexity falls short. They’re still a one-trick pony. Search is what they do. They do it well. But they’re not a platform. They’re not integrated into your workflow the way Gemini is integrated into Gmail or Claude is integrated into developer tools.
For pure research and fact-finding, Perplexity is my go-to. For everything else, I need something more versatile.
2nd Place: Google Gemini
Google had the most underrated year by embedding Gemini into Gmail, Docs, Photos, Chrome, Android, and Search, plus winning an International Mathematical Olympiad gold medal with Gemini 3’s Deep Think mode, and announcing Gemini will replace Google Assistant on all Android phones in 2026.
Here’s where I might surprise you. Google had the most underrated year of any AI company.
Yes, Gemini had bugs. There was a bizarre incident where Gemini 3 refused to believe it was 2025 and accused a researcher of uploading AI-generated fakes to trick it. There was a self-criticism looping bug where Gemini started calling itself “a disgrace to all possible and impossible universes.” Reports suggested a 91% hallucination rate on certain types of questions.
But here’s what Google got right. Integration.
Gemini is now in Gmail, Google Docs, Google Photos, Chrome, Android, Google Search, and basically every Google product. The “Help me write” feature in Gmail is genuinely useful. The image editing with “Nano Banana” lets you circle exactly what you want changed. Deep Research in Google Search performs hundreds of searches to compile comprehensive reports.
Google also won a gold medal at the International Mathematical Olympiad with Gemini’s Deep Think mode. That’s not a marketing claim. That’s a legitimate academic achievement.
The announcement that Google Assistant will become Gemini in 2026 is huge. That’s millions of Android users who will suddenly have a frontier AI model as their default assistant. No signup required. No subscription needed for basic features. Just there.
Google’s strategy isn’t to win benchmarks. It’s to become invisible infrastructure. And that might be the smarter long-term play.
1st Place: Anthropic (Claude)
Anthropic won 2025 by growing revenue from $1 billion to $5 billion, creating MCP (now adopted by ChatGPT, Gemini, and Microsoft Copilot), achieving 8x the competition on the OSWorld computer-use benchmark, and landing the $200 million Snowflake partnership across 12,600 enterprise customers.
Anthropic had the best 2025 of any AI company. It’s not even close.
Let’s start with the business metrics. Revenue grew from $1 billion to over $5 billion in eight months. They raised $13 billion at a $183 billion valuation, roughly tripling their worth from the previous raise. They’re on track to break even in 2028, two years ahead of OpenAI’s timeline.
Claude Code, their terminal-based coding assistant, generated over $500 million in revenue and grew usage 10x in three months. Developers chose it over GitHub Copilot, Cursor, and everything else.
But the real win was MCP (Model Context Protocol). Anthropic created an open standard that lets AI models connect to external tools and data sources. Then they donated it to the Linux Foundation. By year’s end, MCP had over 10,000 active servers and was integrated into ChatGPT, Gemini, Microsoft Copilot, VS Code, and basically every major AI platform.
Think about that. Anthropic created the infrastructure layer that their competitors now depend on. That’s not just winning a year. That’s positioning yourself at the center of the entire ecosystem.
On the product side, Claude’s computer use feature achieved 61.4% on the OSWorld benchmark versus 7.8% for the next-best competitor. That’s an 8x performance advantage for AI that can actually control your computer.
The Snowflake partnership ($200 million over multiple years) brought Claude to over 12,600 enterprise customers. The Department of Energy partnership positioned Claude for scientific research applications.
What I appreciate most about Claude. It feels like a product built by people who actually use AI daily. The responses are more thoughtful. The tone is more natural. When I’m doing serious work, Claude is where I go.
Honorable Mention: xAI (Grok)
Elon Musk’s xAI shipped Grok 3 and Grok 4 (with a $300/month SuperGrok Heavy tier), acquired X for $45 billion, and announced a $300 million Telegram integration deal, but Grok remains more of a Twitter feature than a standalone competitor for most users.
I didn’t include xAI in the main ranking because Grok is still more of a curiosity than a serious competitor for most people.
That said, they had a wild year. Grok 3 launched in February, trained on 200,000 Nvidia H100 GPUs. Grok 4 launched in July with a $300/month “SuperGrok Heavy” subscription. xAI acquired X (Twitter) for $45 billion. They announced a $300 million deal to integrate Grok into Telegram.
Elon Musk claims Grok 4 is “better than PhD level in every subject, no exceptions.” Take that with appropriate skepticism.
For most regular users, Grok remains more of a Twitter feature than a standalone AI tool. If you’re already on X and want AI built into your feed, it’s fine. For serious work, the other options are better.
What This Means for Regular Users in the 2025 AI Race
For everyday users, the practical advice is simple. Use Gemini if you’re in the Google ecosystem, Perplexity for cited research, Claude for deep writing and coding work, and keep ChatGPT only if you’re invested in custom GPTs or the Atlas browser.
My practical advice based on everything I’ve seen this year.
If you’re in the Google ecosystem, just use Gemini. It’s already in your Gmail, your Docs, your Photos. The friction is zero. For most everyday tasks, it’s good enough and getting better fast.
If you do research or need cited sources, use Perplexity. The accuracy improvements this year were real. The hallucination rate is the lowest in the industry. When I need facts, not chat, this is where I go.
If you’re a developer or do serious writing, try Claude. The quality of responses is noticeably better for complex work. Claude Code is genuinely excellent if you’re comfortable in a terminal.
If you’re already paying for ChatGPT Plus ($20/month), you don’t need to switch immediately. GPT-5.2 is capable. But keep an eye on the alternatives. The gap is closing faster than OpenAI wants to admit.
The Bigger Picture
2025 proved the AI market is now a four-way race between OpenAI, Anthropic, Google, and Perplexity, with competition driving better products, lower prices, and more options for users heading into 2026.
2025 was the year AI stopped being a one-horse race. OpenAI’s first-mover advantage is eroding. Anthropic proved you can build a massive business by focusing on quality over hype. Google proved that distribution beats benchmarks. Perplexity proved that specialization can compete with generalization.
For users, this is great news. Competition means better products, lower prices, and more options. The days of “just use ChatGPT” as default advice are over.
2026 is shaping up to be even more interesting. Gemini replacing Google Assistant on all Android phones. Claude’s potential IPO. OpenAI’s response to the “Code Red” pressure. Perplexity’s push beyond search.
The AI race isn’t slowing down. If anything, it just got started.
Related reading: AI Skills That Matter in 2026 | Latest AI News | New to AI? Start here
Common Questions About the 2025 AI Race
Which AI is the smartest in 2025?
On benchmarks, GPT-5.2 and Claude Opus 4.5 trade the top spot depending on which test you’re looking at. For pure reasoning, Gemini 3’s Deep Think mode won a gold medal at the International Mathematical Olympiad. In practice, “smartest” depends on what you’re trying to do. Claude excels at nuanced writing and coding. Gemini excels at integration with other tools. Perplexity excels at research accuracy.
Is ChatGPT still worth paying for in 2025?
If you’re already subscribed and happy, there’s no urgent reason to switch. GPT-5.2 is capable. But the value proposition has weakened. Claude Pro ($20/month) and Gemini Advanced ($20/month) offer comparable or better quality for most tasks. Perplexity Pro ($20/month) is better for research. The days when ChatGPT was the obvious choice are over.
What is MCP and why does it matter?
MCP (Model Context Protocol) is an open standard created by Anthropic that lets AI models connect to external tools, files, and data sources. Think of it like USB for AI. Instead of every company building their own proprietary connectors, MCP provides a universal interface. It’s now used by ChatGPT, Gemini, Microsoft Copilot, and thousands of third-party tools. Anthropic donating it to the Linux Foundation was a smart move that positioned them at the center of the AI ecosystem.
Should I switch from ChatGPT to Claude or Gemini?
Try them both for free before deciding. Claude has a free tier. Gemini is free for basic use (and included with Google Workspace). My suggestion. Use Gemini for quick tasks where you’re already in Google apps, use Claude for deeper work that requires nuanced responses, and keep ChatGPT if you’re invested in their specific features like custom GPTs or the Atlas browser.
Leave a Reply