GPT-5.2 vs Gemini 3 — How they compare

OpenAI’s latest AI model GPT-5.2 is here. But how does it compare to its biggest competitor, Google’s Gemini 3?

The ChatGPT creator launched the GPT-5.2 on Thursday, and it’s currently rolling out to paid ChatGPT customers, with wider access to follow. It comes on the heels of a leaked memo, in which OpenAI CEO Sam Altman declared “code red” for the AI company. OpenAI is facing steeper competition from Google after the release of its latest AI model, Gemini 3. Google’s “most intelligent” AI model yet was surging past its AI competitors in various benchmark tests and on the AI leaderboards at LMArena. 

In addition to that, a post from former Googler Deedy Das claimed that in the two weeks since Gemini 3 had launched, OpenAI lost nearly 6 percent of its visitors.

So, how does OpenAI’s GPT-5.2 stack up against Google’s Gemini 3? So far, the models seem evenly matched on benchmark performance, pricing, and other comparisons, with both models gaining an edge in specific areas.

First, it should be noted that GPT-5.2 just came out, and we’ll know more in the days ahead. Unsurprisingly, OpenAI claims that GPT-5.2 excels over other AI models when it comes to “professional knowledge work.” OpenAI describes this as GPT-5.2 being “better at creating spreadsheets, building presentations, writing code, perceiving images, understanding long contexts, using tools, and handling complex, multi-step projects.”

GPT-5.2 vs. Gemini 3: LMArena leaderboards

Based on the more neutral LMArena leaderboards, GPT-5.2 appears to be impressing users in at least one key area: web development.

The leaderboard has the higher-end GPT-5.2-high in second place, behind Claude Opus 4.5. Gemini 3 Pro is in the fourth spot, and the base GPT-5.2 model is in sixth. So, for web development tasks, it appears GPT-5.2 excels above Gemini 3. Though, again, these are early days.

So far, GPT-5.2 is unranked on the rest of LMArena’s leaderboards.

However, GPT-5.2 has a steep hill to climb. Various versions of Gemini 3 currently top the leaderboards for text, vision, text to image, image edit, and search, with the closely related Google Veo 3 models topping the remaining text to video and image to video categories. Gemini 3 also tops the overall leaderboard, where GPT-5.2 remains unranked.

GPT-5.1-high is currently ranked sixth in the overall rankings, which explains why Altman might be seeing red.

GPT-5.2 vs. Gemini 3: Benchmark tests

At this time, we can’t independently verify the benchmark results OpenAI has reported for GPT-5.2, and third-party rankings from companies like ScaleAI so far don’t include the new model. However, looking at the benchmark results reported by Google and OpenAI, we can make some comparisons.

• SWE-bench verified: Google says Gemini 3 scored 76.2 percent, while OpenAI says GPT-5.2 scored 80 percent

• Humanity’s Last Exam (HLE), no tools: Google says Gemini 3 scored 37.5 percent; OpenAI says GPT-5.2 scored 34.5 percent

• GPQA Diamond: Google says Gemini 3 scored 91.9 percent; OpenAI says GPT-5.2 scored 92.4 percent.

• AIME 2025, no tools: Google says Gemini scored 95 percent without tools; OpenAI says GPT-5.2 scored 100 percent without tools.

• MMMLU: Google says Gemini 3 scored 91.8 percent; OpenAI says GPT-5.2 scored 89.6 percent.

Overall, it’s a mixed bag for both models on benchmark performance — you win some, you lose some. You can check out the full benchmark performance at OpenAI and Google.

GPT-5.2 vs. Gemini 3: Features and availability

Gemini 3 and GPT-5.2 are both series of models, which are then integrated into the Gemini and ChatGPT platforms. And these platforms have different capabilities. For instance, Gemini lets users create both AI-generated photos and videos. ChatGPT lets users create images, but users need to use the Sora app for AI videos. Professional users can also access the new models via Google or OpenAI’s API and enterprise systems.

Google does have the edge in one respect: Gemini 3 is integrated into more products, giving users more ways to access the new reasoning model. In addition to the Gemini and Google apps, Gemini 3 powers Google AI Mode. Google has also integrated Gemini 3 into various Google AI Studio projects and tools like NotebookLM.

Ultimately, you can do more with Gemini 3 than you can with GPT-5.2, even if it turns out GPT-5.2 performs better.

GPT-5.2 vs. Gemini 3: Pricing

Price-wise, both AI models are pretty even. For $20 per month, users get expanded access to the GPT-5.2 models within OpenAI’s ChatGPT Plus plan (the Pro plan costs $200/month). Google also charges $20 per month for its Google AI Pro plan (and $249.99/month for Google AI Ultra, which comes with Google cloud storage). The two are also pretty evenly priced when it comes to API access:

Our recommendation for AI users? Try the two AI models out yourself and see which one best fits your needs and workflow.

Disclosure: Ziff Davis, Mashable’s parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.