XNews

OpenAI releases GPT-5.2 and Code Red response

OpenAI releases GPT-5.2 and Code Red response

OpenAI released GPT-5.2 and announced an internal Code Red to prioritize ChatGPT improvements. - GPT-5.2 Thinking reported large benchmark gains across science, math and coding - SWE-Bench software engineering: 55.6% (from 50.8%) - GPQA Diamond science: 92.4% (from 88.1%) - CharXiv scientific reasoning: 82.1% (from 67.0%) - FrontierMath overall: 40.3% (from 31.0%); Tier 4: 14.6% - AIME 2025 math: 100% (from 94.0%) - ARC-AGI 1: 86.2% (from 72.8%); ARC-AGI 2: 52.9% (from 17.6%) - GDPval knowledge-work metric: 70.9% (from 38.8%) - OpenAI announced three GPT-5.2 variants: Instant, Thinking, Pro - Sam Altman circulated a Code Red memo pausing ad, Agent, Pulse projects to accelerate reasoning - OpenAI plans rollout to paid users this week and an 'adult mode' and age-prediction feature next quarter

Key insights

GPT-5.2 Thinking posts large benchmark improvements

OpenAI announced GPT-5.2 Thinking with substantial numeric gains on multiple benchmarks: SWE-Bench 55.6% (from 50.8%), GPQA Diamond 92.4% (from 88.1%), CharXiv 82.1% (from 67.0%), FrontierMath 40.3% overall (from 31.0%) and AIME 2025 perfect 100% (from 94.0%). Abstract reasoning gains include ARC-AGI 1 at 86.2% (from 72.8%) and ARC-AGI 2 at 52.9% (from 17.6%). OpenAI also reports GDPval knowledge-work improvement to 70.9% (from 38.8%). These metrics were highlighted as central claims in the GPT-5.2 release materials and social posts announcing the model.

OpenAI shifts priorities under 'Code Red' to accelerate ChatGPT

An internal Sam Altman 'Code Red' memorandum directed OpenAI to prioritize ChatGPT improvements and to pause several commercial initiatives. The memo listed five priorities: improve image generation, strengthen model reasoning and dialogue stability, win rankings, increase speed and stability, and reduce spurious refusals. It announced an imminent new reasoning model, paused the ad business, Agent automation projects and Pulse personalized reports, and set a short urgent timeline for releases and engineering focus.

GPT-5.2 product architecture: Instant, Thinking, Pro

OpenAI positioned GPT-5.2 as a family addressing distinct workflows: GPT-5.2 Instant for fast everyday retrieval and light tasks; GPT-5.2 Thinking for deep multi-step reasoning, long-context and knowledge-work tasks; GPT-5.2 Pro as a highest-compute tier for difficult research and decision problems. The company emphasized the Thinking variant as the core update targeting professional work and long-horizon agents.

Commercial rollout, monetization and safety controls announced

OpenAI said GPT-5.2 will start rolling to paid users this week. The company indicated plans for an 'adult mode' and an automated age-prediction feature to restrict under-18 content, with adult-mode planned for Q1 next year. Sam Altman told staff the Code Red status is expected to end by January when the roadmap completes prioritized work.

OpenAI frames GPT-5.2 as a push to reclaim leadership vs. Gemini and Claude

OpenAI described GPT-5.2 as a targeted response to competitor releases (Google Gemini 3 and Anthropic Claude Opus), emphasizing professional knowledge-work reliability and long-context reasoning rather than general-chat improvements. Public materials and commentary positioned GPT-5.2 as engineered to outperform peers on key work-oriented benchmarks and agent readiness.

Claims of human-level task parity and agent orientation

OpenAI and commentators stated GPT-5.2 outperforms human experts on roughly 70% of real job tasks, is optimized for long-horizon agent operation, and is described as 'built for real work'. Observers highlighted the model's focus on reducing multi-step failure modes for agent deployments and knowledge-worker substitution metrics (GDPval).

Source posts