OpenAI Launches GPT‑5.2: SOTA Model for Professional Work and Agentic Workflows

OpenAI launched GPT‑5.2 (Instant, Thinking, Pro), a major upgrade in professional capability with state-of-the-art results across knowledge work, coding, long-context reasoning, tool use, and vision. It improves factuality, handles up to 256k-token contexts more accurately, and executes complex workflows with better tool reliability and lower latency options. Rolling out to paid ChatGPT plans and the API, GPT‑5.2 introduces new pricing, model names, and safety enhancements.

Key Points

GPT‑5.2 (Instant, Thinking, Pro) delivers state-of-the-art performance in professional knowledge work, coding, long-context reasoning, tool use, and vision.
On GDPval, GPT‑5.2 Thinking beats or ties industry professionals 70.9% of the time, with significant gains in spreadsheet/presentation generation and coding (SWE-Bench Pro 55.6%).
Long-context reasoning leads on OpenAI MRCRv2 (near-100% at 256k for 4-needle), with a new /compact endpoint extending effective context for long-running, tool-heavy workflows.
Tool calling is far more reliable (Tau2-bench Telecom 98.7%), with improved low-latency performance and stronger end-to-end agentic workflows.
Available now in ChatGPT paid plans and the API with new pricing (gpt‑5.2: $1.75/M input, $14/M output), safety upgrades for sensitive content, and maintained support for GPT‑5.1.

Sentiment

The overall sentiment of the Hacker News discussion is largely skeptical and critical towards OpenAI's GPT-5.2 release. While some acknowledge the impressive benchmark improvements, a significant portion of the community expresses distrust in OpenAI's marketing, questions the real-world utility and practical speed of the model, and raises concerns about the increased pricing and perceived lack of transparency. There's a strong undercurrent of 'show me, don't tell me' regarding the model's capabilities, with many users pointing out inconsistencies and potential marketing ploys.

In Agreement

The benchmark improvements, particularly for ARC AGI v2, SWE Verified, and AIME 100% without tools, are noted as impressive.
The release is seen as a positive outcome of competition, specifically from Gemini 3 Pro, pushing OpenAI to ship models faster.
The new model, especially gpt-5.2 with xhigh reasoning, offers significantly better cost-per-task performance on benchmarks like ARC AGI v2 compared to previous models.
Some users find the voice chat quality of ChatGPT (or Claude) to be excellent and a killer feature.
The Pro model, despite its high cost, is considered valuable by some for solving complex problems on the first try, saving significant time.
There is a belief that AI development has not hit an 'S-curve' wall and will continue to advance exponentially, eventually mimicking all human behavior.

Opposed

Users immediately identified multiple factual errors in OpenAI's promotional motherboard image, casting doubt on the 'better vision' claim and suggesting it's misleading.
There is widespread skepticism regarding the validity and real-world applicability of benchmarks, with concerns about 'benchmark saturation,' 'gaming the benchmark,' and training models on test data.
The lack of direct comparisons to competitor models in OpenAI's announcement is interpreted as a sign that GPT-5.2 might be underperforming against rivals like Claude Opus or Gemini 3 Pro.
Many users express concerns about the significant 40% price increase for API models, particularly the 'exorbitant' pricing of the Pro model, which they feel is not justified by marginal performance gains.
The model is criticized for being too slow in practical use, taking too long to generate responses for tasks that competitors handle faster, and for producing 'laughable' or unreliable results in complex coding or long-horizon tasks despite high benchmark scores.
Skepticism exists about whether GPT-5.2 is a truly new, pre-trained model or merely an optimization/tweaking of existing models, rushed out due to competitive pressure, especially given the rapid release after 5.1 and rumors of OpenAI's pre-training challenges.
The August 2025 knowledge cutoff for a point release is questioned, as it implies a recent pre-training run which contradicts industry rumors about OpenAI's difficulties in full-scale pre-training.
OpenAI is criticized for lacking transparency regarding training specifics and for not even proofreading its own press release, indicating a lack of internal integration of their own models for quality control.
Frustration is expressed over OpenAI's 100% AI-based customer support, which cannot escalate issues to human agents or provide information on early access programs like fine-tuning.
Users anticipate that the quality of ChatGPT will degrade over time due to throttling, a common complaint with previous OpenAI models.