Google Gemini 3 Flash vs OpenAI GPT-5.2: Benchmarking the Next Generation of AI Models

Daily Technology

18/12/2025

Introduction

Google released Gemini 3 Flash in June 2025. The new model became the default inside the Gemini app the same day and now powers parts of Google Search plus business tools. The launch pits the system against OpenAI's GPT-5.2, since both aim for strong performance, multimodal skills and easy scaling. This article compares the two using standard benchmarks but also explains what the technical gaps mean for companies and for everyday users.

Model Performance on Public Benchmarks

Benchmark scores still decide which model looks best. On “Humanity's Last Exam,” a wide ranging test that checks knowledge across dozens of fields, Gemini 3 Flash reached 33.7 %, just under GPT-5.2 at 34.5 %. Gemini 3 Pro topped both at 37.5 %, while the older Gemini 2.5 Flash stopped at 11 %. On MMMU-Pro, a test that mixes images, charts as well as reasoning, Gemini 3 Flash scored 81.2 %, the highest result published so far showing clear progress on tasks that mix pictures with text.

Speed, Efficiency and Cost Structure

Google states that Gemini 3 Flash finishes work three times faster than Gemini 2.5 Pro or needs 30 % fewer tokens for the same job. The listed price is $0.50 for every million input tokens and $3.00 for every million output tokens. That rate is slightly above the cost of 2.5 Flash - yet the jump in speed also the drop in token use balance the difference for large, repeatable jobs. OpenAI has not published fresh efficiency numbers for GPT-5.2 - a direct cost comparison is not yet possible.

Multimodal and Functional Capabilities

Gemini 3 Flash accepts video, images, audio and code in a single session - it returns answers that contain tables or pictures instead of plain text. A user can upload a short video of a video game next to ask for strategy tips or hand draw a rough diagram and request a polished version. The model is live on Vertex AI, Gemini Enterprise plus the standard Google API - developers can build prototypes without special access. GPT-5.2 also targets business use and higher message limits, but OpenAI has not released matching benchmark figures for image, video or audio tasks.

Market Impact but also Industry Deployment

Large companies started using Gemini 3 Flash within weeks of release. JetBrains weaves it into coding software, Figma adds it to design review steps besides Harvey, a legal tech firm, relies on it for contract checks. Each case shows that the model handles scale and keeps output stable. OpenAI answers Google's pace with rapid internal cycles as well as frequent updates. The back-and-forth pushes both firms to ship new versions and to set tougher targets for the rest of the field.

Conclusion

Google Gemini 3 Flash or OpenAI GPT-5.2 sit at the leading edge of language or multimodal work. Test scores place them almost side by side on knowledge tasks, while Gemini 3 Flash moves ahead on certain image-and-text benchmarks and on raw speed. Enterprises that need reliable, flexible AI will find strong value in either system. Ongoing tests also open numbers will stay vital for anyone who must pick a model today or track progress tomorrow.

2025-08-09

OpenAI's GPT-5 Learns to Say 'I Don't Know,' Ushering in Era of Honest AI

OpenAI's new GPT-5 model introduces humility, admitting when it doesn't know answers and reducing fabricated information, aiming to build trust in AI.