The AI race just accelerated—and faster than anyone expected. OpenAI has officially released GPT-5.2, barely a month after GPT-5.1, signaling a clear sense of urgency inside the company. With Google’s Gemini 3 rapidly climbing benchmarks and Anthropic’s Claude Opus 4.5 winning over developers, OpenAI needed a decisive response. GPT-5.2 is that response: faster, smarter, better at reasoning, and clearly aimed at regaining benchmark dominance.

What’s unusual isn’t just the speed of this release—it’s the context. In early December, Sam Altman reportedly issued a “code red” internally after ChatGPT’s traffic dipped and Gemini gained traction. Advertising rollouts, side projects, and experimental features were paused so teams could focus on one thing: releasing the strongest model possible, as quickly as possible.

And on paper, GPT-5.2 delivers.

GPT-5.2 Comes in Three Versions

Just like earlier generations, GPT-5.2 is available in three modes:

• GPT-5.2 Instant – fastest responses for everyday use
• GPT-5.2 Thinking – multi-step reasoning and complex tasks
• GPT-5.2 Pro – maximum precision for expert-level work

OpenAI claims GPT-5.2 surpasses professional human performance on 70.9% of evaluated work tasks, from code and spreadsheets to business presentations. Although OpenAI’s internal metrics should always be viewed critically, independent benchmarks confirm clear improvements in reasoning and coding.

The AI War Is Officially On

Google is stronger than ever. Gemini 3, released in late November, took the lead on many evaluation platforms—especially LMArena, the go-to site for comparing LLMs. Google now touts 650 million monthly users across its Gemini ecosystem, with deep integration into Google Cloud, Vertex AI, and its code assistant tools.

OpenAI counters with an even bigger figure: 800 million weekly ChatGPT users, claiming nearly 70% of the global conversational AI market. But internal metrics reportedly show declining usage, which contributed to the urgent push for GPT-5.2.

Meanwhile, Anthropic is becoming the developers’ favorite. Its new Claude Opus 4.5 leads in hardcore coding benchmarks like SWE-Bench Verified (80.9%) and continues to gain traction with professionals.

OpenAI now finds itself squeezed between:

• Google → massive cloud infrastructure + unmatched financial resources
• Anthropic → developer-first, highly aligned, top coding accuracy

GPT-5.2 is OpenAI’s attempt to retake the lead.

GPT-5.2’s Benchmark Performance: Impressive Across the Board

Huge Leap in Professional Task Accuracy

On OpenAI’s internal GDPval test (44 professional skills), GPT-5.2 hits 70.9%, up from 38.8% on GPT-5. The Thinking version reportedly equals or exceeds human professionals while being:

• 11× faster
• at <1% of the cost per task

Massive Upgrade in Reasoning

GPT-5.2 Thinking makes a dramatic jump on reasoning benchmarks:

• ARC-AGI-2: 52.9% (vs. 17.6% on GPT-5.1)
• ARC-AGI-1: 90.5% with GPT-5.2 Pro — the first model to break 90%

These tests simulate true abstract reasoning, not memorization—so the improvement is meaningful.

Code & Math: Strong Gains

• SWE-Bench Pro: 55.6% (vs. 50.8% for GPT-5.1)
• AIME 2025: 100%
• FrontierMath: 40.3% on research-level problems

OpenAI also reports:

• 38% fewer factual errors
• 30% fewer hallucinations

But the Competition Still Leads in Key Areas

GPT-5.2 isn’t a clean sweep. Its rivals hold major advantages:

Anthropic’s Claude Opus 4.5

• SWE-Bench Verified: 80.9% (slightly ahead of GPT-5.2’s ~80%)
• Terminal-Bench 2.0: 59.3% (still best for command-line tasks)

Claude remains the preferred model for developers needing extremely reliable coding assistance.

Google’s Gemini 3 Pro

Still competitive—and sometimes ahead:

• GPQA Diamond: 91.9% (vs. 92.4% for GPT-5.2 Thinking)
• AIME 2025: 95% with code execution enabled
• Excels in vision and multimodal reasoning

Evaluations from LMArena still show Gemini 3 leading in:

• autonomous coding
• visual comprehension
• several professional tasks

GPT-5.2 closes the gap, but doesn’t dominate.

Pricing: More Power, Higher Cost

OpenAI has increased prices across the board compared to GPT-5.1.

GPT-5.2 Thinking (API Pricing)

• Input: $1.75 per million tokens
• Output: $14 per million tokens

Previously: $1.25 / $10.

GPT-5.2 Pro

• Input: $21
• Output: $168

This is roughly 40% more expensive than GPT-5 Pro.

OpenAI says the increase reflects:

• higher computational cost of multi-step reasoning
• deeper internal chain-of-thought processing
• improved task success rate (reducing total cost per task)

When compared to competitors:

• Gemini 3 Pro: $2 input / $12 output
• Claude Opus 4.5: $5 input / $25 output

GPT-5.2 sits in the middle.

Availability and API Access

GPT-5.2 launched on December 11, 2025 and is currently available only to:

• ChatGPT Plus users (€23/mo)
• ChatGPT Pro users (€229/mo)

Free users must wait—OpenAI hasn’t shared a timeline.

API model IDs available now

• gpt-5.2 → Thinking
• gpt-5.2-chat-latest → Instant
• gpt-5.2-pro → Pro

OpenAI confirms that GPT-5.1, GPT-5, and GPT-4.1 will remain available, with no deprecation planned.

A specialized Codex-optimized version of GPT-5.2 is also coming soon to boost developer performance.

Conclusion:

GPT-5.2 is one of OpenAI’s fastest and most aggressive releases ever—an emergency upgrade intended to counter Google and Anthropic’s rapid progress. It truly improves reasoning, code accuracy, and professional task performance. But it does not decisively dominate the field.

Instead, GPT-5.2 marks the beginning of a new phase:
an AI race where each company is willing to move faster, launch sooner, and iterate more aggressively than ever before.

For users and developers, the outcome is clear:
AI models are evolving at record speed, and 2026 is poised to be even more competitive.