OpenAI Unveils GPT-5.2 As Competition With Google’s Gemini Intensifies

After weeks of industry speculation and an internal “Code Red”, OpenAI has formally introduced GPT-5.2, its latest frontier model aimed at reinforcing its leadership amid tightening competition with Google’s Gemini 3.

The update comes barely a month after GPT-5.1 and lands at a moment when the competitive landscape is shifting fast. Earlier this month, reports suggested CEO Sam Altman urged teams to prioritise improving the ChatGPT experience over new commercial initiatives such as ads, following concerns around declining traffic and growing user migration toward Google.

Despite this backdrop, OpenAI maintains that GPT-5.2 is not a reactionary launch. Executives emphasised during the media briefing that the model represents steady progress towards more capable reasoning systems, not a direct counter to Gemini.

What's new in GPT-5.2 

OpenAI is positioning GPT-5.2 as its most capable model series so far, particularly for developers and professional knowledge workers. The model line is available in three variants: Instant, Thinking and Pro, each optimised for different workloads.

Fidji Simo, Chief Product Officer, described the release as designed to “unlock even more economic value for people,” highlighting improvements in spreadsheets, presentations, code generation, image perception, handling long context, and multi-step planning.

OpenAI added, “Overall, GPT-5.2 brings significant improvements in general intelligence, long-context understanding, agentic tool-calling, and vision, making it better at executing complex, real-world tasks end-to-end than any previous model.”

These improvements align with OpenAI’s broader strategy to target enterprise and developer ecosystems, positioning GPT-5.2 as the foundation for production-grade AI applications.

Benchmark Gains:

Across benchmarks shared during the launch, GPT-5.2 Thinking shows marked improvements over GPT-5.1 in reasoning, scientific problem solving, advanced maths, and long-context logic.

Key benchmark highlights include:

• Knowledge Work Tasks: 70.9% vs 38.8%

• SWE-Bench Pro: 55.6% vs 50.8%

• GPQA Diamond Science Questions: 92.4% vs 88.1%

• ARC-AGI-2 Abstract Reasoning: 52.9% vs 17.6%

• AIME 2025 (no tools): 100% vs 94%

Aidan Clark, Research Lead, explained why maths remains a critical signal.  “These are all properties that really matter across a wide range of different workloads. Things like financial modelling, forecasting, and doing an analysis of data.”

With stronger logic consistency and fewer stepwise errors, GPT-5.2 Thinking responses reportedly contain 38% fewer errors than its predecessor. For enterprises deploying these systems for modelling, audits, research, or automation, reliability becomes a competitive differentiator.

Positioning Against Google’s Gemini 3

The launch comes at a time when Google’s Gemini 3 is deeply integrated into its ecosystem, from Maps and BigQuery to managed MCP servers — making multimodal and agentic workflows easier to deploy for businesses in the Google Cloud universe.

Google has emphasised that Gemini 3 Pro “significantly outperforms 2.5 Pro on every major AI benchmark” and is a step forward toward AGI. Its Deep Think mode directly targets the reasoning domain, the same territory OpenAI is strengthening with GPT-5.2.

Sam Altman recently told CNBC that Gemini 3’s actual impact was “lesser than he had anticipated”, even though the competitive pressure was real enough to trigger the internal Code Red earlier in December.

Altman added that he expects OpenAI to exit the Code Red state by January.

A High-Stakes Infrastructure Bet

While GPT-5.2 consolidates OpenAI’s recent advancements, it arrives during increased scrutiny over the company’s infrastructure commitments, estimated at $1.4 trillion over the next few years. That scale-out was planned when OpenAI enjoyed first-mover advantage, a position now contested by Google and other frontier-model developers.

Reasoning-heavy models like Thinking and Deep Research carry high compute costs. Running these systems at scale risks creating a cycle where the race to lead benchmarks demands ever-increasing infrastructure outlay.

During the briefing, Simo argued that efficiencies are also improving: “You are getting, today, a lot more intelligence for the same amount of compute and the same amount of dollars as you were a year ago.”

Still, OpenAI’s compute expenditure, reportedly paid mostly in cash rather than cloud credits, signals how aggressively the company is pushing its infrastructure roadmap.

Gaps and What Comes Next

One notable omission from this release is a new image generator. Altman’s Code Red memo highlighted image generation as a renewed priority, especially after Google’s “Nano Banana” (Gemini Flash Image) models gained momentum.

Google recently followed up with Nano Banana Pro, now integrated across several productivity tools, including Mixboard for automated presentations.

OpenAI hinted at another upcoming model in January with better image capabilities, improved speed, and a more customisable user experience, though it declined to confirm timelines.

Meanwhile, GPT-5.2 is rolling out to paid ChatGPT plans and API customers, alongside new safety measures for mental health uses and teenage verification, areas that saw minimal emphasis at the launch.

GPT-5.2 is less of a reinvention and more of an incremental reinforcement, a consolidation of GPT-5’s routing architecture and GPT-5.1’s conversational warmth and agentic enhancements. But its timing makes it a critical signal: OpenAI intends to stay in the lead on reasoning workloads, the very arenas enterprises depend on for automation, audits, modelling, and decision support.

The competitive race with Google has moved beyond user interfaces and chatbots. It is now about who can deliver reliable, production-ready intelligence at scale. In that race, GPT-5.2 is OpenAI’s clearest attempt yet to reassert momentum.