OpenAI Launches GPT-5.2: New AI Model Beats Gemini 3

December 12, 2025 – OpenAI has rolled out GPT-5.2, its latest frontier AI model, just one day after an intense internal push to accelerate development. The new release, launched on December 11, 2025, delivers significant advancements in spreadsheets, presentations, image perception, coding, and long-context understanding, while reclaiming the top position on several key benchmarks.

The launch follows reports of a “code red” directive from CEO Sam Altman, aimed at rapidly closing the lead held by Google’s recently released Gemini 3 series. Early independent evaluations show GPT-5.2 outperforming its predecessors and narrowly surpassing Gemini 3 on multiple metrics, signaling that OpenAI has reclaimed the edge in the race for the world’s most capable AI.

In this comprehensive report from WorldReport.press, we break down what GPT-5.2 brings to the table, the benchmarks it dominates, and what the accelerated timeline means for the future of artificial intelligence.

Major Improvements in Real-World Professional Tasks

OpenAI emphasized that GPT-5.2 excels at practical workflows that professionals and developers rely on daily:

• Spreadsheets & Presentations — The model can now read, analyze, and generate complex Excel files, Google Sheets, and PowerPoint/Google Slides decks with near-perfect accuracy, handling multi-sheet financial models, pivot tables, and chart creation in seconds.
• Image Perception — Enhanced vision capabilities allow GPT-5.2 to interpret screenshots, handwritten notes, diagrams, and even low-quality photos more precisely than any previous OpenAI model.
• Coding — Developers report that GPT-5.2 solves complex engineering problems faster and with fewer errors, excelling at multi-file projects and debugging.
• Long-Context Understanding — With a context window exceeding 1 million tokens (exact size not yet disclosed), the model maintains coherence across extremely long documents, codebases, or conversations.

Benchmark Dominance: SWE-Bench Pro, GPQA Diamond, and More

OpenAI claims GPT-5.2 sets new records on several high-difficulty evaluations:

• SWE-Bench Pro (real-world software engineering tasks): 48.7% success rate, surpassing Gemini 3’s previous high of 46.1% and becoming the first model to break the 48% barrier.
• GPQA Diamond (graduate-level science questions): 68.4% accuracy, narrowly beating the prior leader and demonstrating superior reasoning in physics, chemistry, and biology.
• Additional top scores on math benchmarks (MATH, GSM8K), coding challenges (HumanEval, MBPP+), and multimodal evaluations.

These results position GPT-5.2 as the current state-of-the-art model across a broad range of academic and professional tasks.

The “Code Red” Directive: OpenAI’s Rapid Acceleration

According to internal sources, Sam Altman issued a “code red” directive in late November 2025, mobilizing engineering teams to prioritize the next major release. The goal: quickly close the gap created by Google’s Gemini 3 launch earlier this year.

The compressed timeline—from internal testing to public rollout in just weeks—highlights OpenAI’s determination to maintain leadership in the fast-moving AI race. GPT-5.2 marks the first major model since the GPT-5 family debuted in mid-2025, and the rapid iteration suggests more frequent updates could become the norm.

Availability and Pricing

GPT-5.2 is now rolling out to ChatGPT Plus, Team, and Enterprise users, with API access available immediately for developers. Pricing remains unchanged for now, though OpenAI has hinted at potential tiered options for heavy users in the coming weeks.

What This Means for the AI Landscape

With GPT-5.2 reclaiming the top spot on several benchmarks, the competition between OpenAI and Google has reached new heights. Google’s Gemini 3 had held the lead for several months, but OpenAI’s swift response demonstrates that neither company is willing to fall behind for long.

For businesses, developers, researchers, and everyday users, the key takeaway is clear: the world’s most powerful AI just got even better—and it’s available today.