Johannes Heidecke

Aaron Jaech, Adam Kalai, Adam Lerer et al. · openai

The o1 model series is trained with large-scale reinforcement learning to reason using chain of thought. These advanced reasoning capabilities provide new avenues for improving the safety and robustness of our models. In particular, our models can reason about our safety policies in context when responding to potentially unsafe prompts, through deliberative alignment. This leads to state-of-the-art performance on certain benchmarks for risks such as generating illicit advice, choosing stereotyped responses, and succumbing to known jailbreaks. Training models to incorporate a chain of thought before answering has the potential to unlock substantial benefits, while also increasing potential risks that stem from heightened intelligence. Our results underscore the need for building robust alignment methods, extensively stress-testing their efficacy, and maintaining meticulous risk management protocols. This report outlines the safety work carried out for the OpenAI o1 and OpenAI o1-mini models, including safety evaluations, external red teaming, and Preparedness Framework evaluations.

Aaron Hurst, Adam Lerer, Adam P. Goucher et al. · openai

GPT-4o is an autoregressive omni model that accepts as input any combination of text, audio, image, and video, and generates any combination of text, audio, and image outputs. It's trained end-to-end across text, vision, and audio, meaning all inputs and outputs are processed by the same neural network. GPT-4o can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time in conversation. It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50\% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models. In line with our commitment to building AI safely and consistent with our voluntary commitments to the White House, we are sharing the GPT-4o System Card, which includes our Preparedness Framework evaluations. In this System Card, we provide a detailed look at GPT-4o's capabilities, limitations, and safety evaluations across multiple categories, focusing on speech-to-speech while also evaluating text and image capabilities, and measures we've implemented to ensure the model is safe and aligned. We also include third-party assessments on dangerous capabilities, as well as discussion of potential societal impacts of GPT-4o's text and vision capabilities.

Sandhini Agarwal, Lama Ahmad, Jason Ai et al. · openai

We present gpt-oss-120b and gpt-oss-20b, two open-weight reasoning models that push the frontier of accuracy and inference cost. The models use an efficient mixture-of-expert transformer architecture and are trained using large-scale distillation and reinforcement learning. We optimize the models to have strong agentic capabilities (deep research browsing, python tool use, and support for developer-provided functions), all while using a rendered chat format that enables clear instruction following and role delineation. Both models achieve strong results on benchmarks ranging from mathematics, coding, and safety. We release the model weights, inference implementations, tool environments, and tokenizers under an Apache 2.0 license to enable broad use and further research.

Josh Achiam, Steven Adler, Sandhini Agarwal et al. · berkeley, deepmind

We report the development of GPT-4, a large-scale, multimodal model which can accept image and text inputs and produce text outputs. While less capable than humans in many real-world scenarios, GPT-4 exhibits human-level performance on various professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers. GPT-4 is a Transformer-based model pre-trained to predict the next token in a document. The post-training alignment process results in improved performance on measures of factuality and adherence to desired behavior. A core component of this project was developing infrastructure and optimization methods that behave predictably across a wide range of scales. This allowed us to accurately predict some aspects of GPT-4's performance based on models trained with no more than 1/1,000th the compute of GPT-4.

Aaditya Singh, Adam Fry, Adam Perelman et al. · berkeley, mila

This is the system card published alongside the OpenAI GPT-5 launch, August 2025. GPT-5 is a unified system with a smart and fast model that answers most questions, a deeper reasoning model for harder problems, and a real-time router that quickly decides which model to use based on conversation type, complexity, tool needs, and explicit intent (for example, if you say 'think hard about this' in the prompt). The router is continuously trained on real signals, including when users switch models, preference rates for responses, and measured correctness, improving over time. Once usage limits are reached, a mini version of each model handles remaining queries. This system card focuses primarily on gpt-5-thinking and gpt-5-main, while evaluations for other models are available in the appendix. The GPT-5 system not only outperforms previous models on benchmarks and answers questions more quickly, but -- more importantly -- is more useful for real-world queries. We've made significant advances in reducing hallucinations, improving instruction following, and minimizing sycophancy, and have leveled up GPT-5's performance in three of ChatGPT's most common uses: writing, coding, and health. All of the GPT-5 models additionally feature safe-completions, our latest approach to safety training to prevent disallowed content. Similarly to ChatGPT agent, we have decided to treat gpt-5-thinking as High capability in the Biological and Chemical domain under our Preparedness Framework, activating the associated safeguards. While we do not have definitive evidence that this model could meaningfully help a novice to create severe biological harm -- our defined threshold for High capability -- we have chosen to take a precautionary approach.

Rahul K. Arora, Jason Wei, Rebecca Soskin Hicks et al.

We present HealthBench, an open-source benchmark measuring the performance and safety of large language models in healthcare. HealthBench consists of 5,000 multi-turn conversations between a model and an individual user or healthcare professional. Responses are evaluated using conversation-specific rubrics created by 262 physicians. Unlike previous multiple-choice or short-answer benchmarks, HealthBench enables realistic, open-ended evaluation through 48,562 unique rubric criteria spanning several health contexts (e.g., emergencies, transforming clinical data, global health) and behavioral dimensions (e.g., accuracy, instruction following, communication). HealthBench performance over the last two years reflects steady initial progress (compare GPT-3.5 Turbo's 16% to GPT-4o's 32%) and more rapid recent improvements (o3 scores 60%). Smaller models have especially improved: GPT-4.1 nano outperforms GPT-4o and is 25 times cheaper. We additionally release two HealthBench variations: HealthBench Consensus, which includes 34 particularly important dimensions of model behavior validated via physician consensus, and HealthBench Hard, where the current top score is 32%. We hope that HealthBench grounds progress towards model development and applications that benefit human health.

Giulio Starace, Oliver Jaffe, Dane Sherburn et al.

We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research. Agents must replicate 20 ICML 2024 Spotlight and Oral papers from scratch, including understanding paper contributions, developing a codebase, and successfully executing experiments. For objective evaluation, we develop rubrics that hierarchically decompose each replication task into smaller sub-tasks with clear grading criteria. In total, PaperBench contains 8,316 individually gradable tasks. Rubrics are co-developed with the author(s) of each ICML paper for accuracy and realism. To enable scalable evaluation, we also develop an LLM-based judge to automatically grade replication attempts against rubrics, and assess our judge's performance by creating a separate benchmark for judges. We evaluate several frontier models on PaperBench, finding that the best-performing tested agent, Claude 3.5 Sonnet (New) with open-source scaffolding, achieves an average replication score of 21.0%. Finally, we recruit top ML PhDs to attempt a subset of PaperBench, finding that models do not yet outperform the human baseline. We open-source our code (https://github.com/openai/preparedness) to facilitate future research in understanding the AI engineering capabilities of AI agents.

Samuel Miserendino, Michele Wang, Tejal Patwardhan et al.

We introduce SWE-Lancer, a benchmark of over 1,400 freelance software engineering tasks from Upwork, valued at \$1 million USD total in real-world payouts. SWE-Lancer encompasses both independent engineering tasks--ranging from \$50 bug fixes to \$32,000 feature implementations--and managerial tasks, where models choose between technical implementation proposals. Independent tasks are graded with end-to-end tests triple-verified by experienced software engineers, while managerial decisions are assessed against the choices of the original hired engineering managers. We evaluate model performance and find that frontier models are still unable to solve the majority of tasks. To facilitate future research, we open-source a unified Docker image and a public evaluation split, SWE-Lancer Diamond (https://github.com/openai/SWELancer-Benchmark). By mapping model performance to monetary value, we hope SWE-Lancer enables greater research into the economic impact of AI model development.

Eric Wallace, Kai Xiao, Reimar Leike et al.

Today's LLMs are susceptible to prompt injections, jailbreaks, and other attacks that allow adversaries to overwrite a model's original instructions with their own malicious prompts. In this work, we argue that one of the primary vulnerabilities underlying these attacks is that LLMs often consider system prompts (e.g., text from an application developer) to be the same priority as text from untrusted users and third parties. To address this, we propose an instruction hierarchy that explicitly defines how models should behave when instructions of different priorities conflict. We then propose a data generation method to demonstrate this hierarchical instruction following behavior, which teaches LLMs to selectively ignore lower-privileged instructions. We apply this method to GPT-3.5, showing that it drastically increases robustness -- even for attack types not seen during training -- while imposing minimal degradations on standard capabilities.

Melody Y. Guan, Manas Joglekar, Eric Wallace et al.

As large-scale language models increasingly impact safety-critical domains, ensuring their reliable adherence to well-defined principles remains a fundamental challenge. We introduce Deliberative Alignment, a new paradigm that directly teaches the model safety specifications and trains it to explicitly recall and accurately reason over the specifications before answering. We used this approach to align OpenAI's o-series models, and achieved highly precise adherence to OpenAI's safety policies, without requiring human-written chain-of-thoughts or answers. Deliberative Alignment pushes the Pareto frontier by simultaneously increasing robustness to jailbreaks while decreasing overrefusal rates, and also improves out-of-distribution generalization. We demonstrate that reasoning over explicitly specified policies enables more scalable, trustworthy, and interpretable alignment.

Tong Mu, Alec Helyar, Johannes Heidecke et al.

Reinforcement learning based fine-tuning of large language models (LLMs) on human preferences has been shown to enhance both their capabilities and safety behavior. However, in cases related to safety, without precise instructions to human annotators, the data collected may cause the model to become overly cautious, or to respond in an undesirable style, such as being judgmental. Additionally, as model capabilities and usage patterns evolve, there may be a costly need to add or relabel data to modify safety behavior. We propose a novel preference modeling approach that utilizes AI feedback and only requires a small amount of human data. Our method, Rule Based Rewards (RBR), uses a collection of rules for desired or undesired behaviors (e.g. refusals should not be judgmental) along with a LLM grader. In contrast to prior methods using AI feedback, our method uses fine-grained, composable, LLM-graded few-shot prompts as reward directly in RL training, resulting in greater control, accuracy and ease of updating. We show that RBRs are an effective training method, achieving an F1 score of 97.1, compared to a human-feedback baseline of 91.7, resulting in much higher safety-behavior accuracy through better balancing usefulness and safety.

Rebecca Soskin Hicks, Mikhail Trofimov, Dominick Lim et al.

Millions of clinicians use ChatGPT to support clinical care, but evaluations of the most common use cases in model-clinician conversations are limited. We introduce HealthBench Professional, an open benchmark for evaluating large language models on real tasks that clinicians bring to ChatGPT in the course of their work. The benchmark is organized around three common use cases central to clinical practice: care consult, writing and documentation, and medical research. Each example includes a physician-authored conversation with ChatGPT for Clinicians and is scored via rubrics written and iteratively adjudicated by three or more physicians across three phases. HealthBench Professional examples were carefully selected for quality, representativeness, and difficulty for OpenAI's current frontier models, to enable continued measurement of progress. Difficult examples for recent OpenAI models were enriched by roughly 3.5 times relative to the candidate pool of 15,079 examples. Additionally, about one-third of examples involve physicians conducting deliberate adversarial testing of models. As a strong baseline, we also collected human physician responses for all tasks (unbounded time, specialist-matched, web access). The best scoring system, GPT-5.4 in ChatGPT for Clinicians, outperforms base GPT-5.4, all other models, and human physicians. We hope HealthBench Professional provides the healthcare AI community a measure to track frontier model progress in real-world clinical tasks and build systems that clinicians can trust to improve care.

Wojciech Zaremba, Evgenia Nitishinskaya, Boaz Barak et al.

We conduct experiments on the impact of increasing inference-time compute in reasoning models (specifically OpenAI o1-preview and o1-mini) on their robustness to adversarial attacks. We find that across a variety of attacks, increased inference-time compute leads to improved robustness. In many cases (with important exceptions), the fraction of model samples where the attack succeeds tends to zero as the amount of test-time compute grows. We perform no adversarial training for the tasks we study, and we increase inference-time compute by simply allowing the models to spend more compute on reasoning, independently of the form of attack. Our results suggest that inference-time compute has the potential to improve adversarial robustness for Large Language Models. We also explore new attacks directed at reasoning models, as well as settings where inference-time compute does not improve reliability, and speculate on the reasons for these as well as ways to address them.

Tyna Eloundou, Alex Beutel, David G. Robinson et al.

Evaluating chatbot fairness is crucial given their rapid proliferation, yet typical chatbot tasks (e.g., resume writing, entertainment) diverge from the institutional decision-making tasks (e.g., resume screening) which have traditionally been central to discussion of algorithmic fairness. The open-ended nature and diverse use-cases of chatbots necessitate novel methods for bias assessment. This paper addresses these challenges by introducing a scalable counterfactual approach to evaluate "first-person fairness," meaning fairness toward chatbot users based on demographic characteristics. Our method employs a Language Model as a Research Assistant (LMRA) to yield quantitative measures of harmful stereotypes and qualitative analyses of demographic differences in chatbot responses. We apply this approach to assess biases in six of our language models across millions of interactions, covering sixty-six tasks in nine domains and spanning two genders and four races. Independent human annotations corroborate the LMRA-generated bias evaluations. This study represents the first large-scale fairness evaluation based on real-world chat data. We highlight that post-training reinforcement learning techniques significantly mitigate these biases. This evaluation provides a practical methodology for ongoing bias monitoring and mitigation.

Miles Wang, Tom Dupré la Tour, Olivia Watkins et al.

Understanding how language models generalize behaviors from their training to a broader deployment distribution is an important problem in AI safety. Betley et al. discovered that fine-tuning GPT-4o on intentionally insecure code causes "emergent misalignment," where models give stereotypically malicious responses to unrelated prompts. We extend this work, demonstrating emergent misalignment across diverse conditions, including reinforcement learning on reasoning models, fine-tuning on various synthetic datasets, and in models without safety training. To investigate the mechanisms behind this generalized misalignment, we apply a "model diffing" approach using sparse autoencoders to compare internal model representations before and after fine-tuning. This approach reveals several "misaligned persona" features in activation space, including a toxic persona feature which most strongly controls emergent misalignment and can be used to predict whether a model will exhibit such behavior. Additionally, we investigate mitigation strategies, discovering that fine-tuning an emergently misaligned model on just a few hundred benign samples efficiently restores alignment.

Alex Beutel, Kai Xiao, Johannes Heidecke et al.

Automated red teaming can discover rare model failures and generate challenging examples that can be used for training or evaluation. However, a core challenge in automated red teaming is ensuring that the attacks are both diverse and effective. Prior methods typically succeed in optimizing either for diversity or for effectiveness, but rarely both. In this paper, we provide methods that enable automated red teaming to generate a large number of diverse and successful attacks. Our approach decomposes the task into two steps: (1) automated methods for generating diverse attack goals and (2) generating effective attacks for those goals. While we provide multiple straightforward methods for generating diverse goals, our key contributions are to train an RL attacker that both follows those goals and generates diverse attacks for those goals. First, we demonstrate that it is easy to use a large language model (LLM) to generate diverse attacker goals with per-goal prompts and rewards, including rule-based rewards (RBRs) to grade whether the attacks are successful for the particular goal. Second, we demonstrate how training the attacker model with multi-step RL, where the model is rewarded for generating attacks that are different from past attempts further increases diversity while remaining effective. We use our approach to generate both prompt injection attacks and prompts that elicit unsafe responses. In both cases, we find that our approach is able to generate highly-effective and considerably more diverse attacks than past general red-teaming approaches.

Robert Korom, Sarah Kiptinness, Najib Adan et al.

We evaluate the impact of large language model-based clinical decision support in live care. In partnership with Penda Health, a network of primary care clinics in Nairobi, Kenya, we studied AI Consult, a tool that serves as a safety net for clinicians by identifying potential documentation and clinical decision-making errors. AI Consult integrates into clinician workflows, activating only when needed and preserving clinician autonomy. We conducted a quality improvement study, comparing outcomes for 39,849 patient visits performed by clinicians with or without access to AI Consult across 15 clinics. Visits were rated by independent physicians to identify clinical errors. Clinicians with access to AI Consult made relatively fewer errors: 16% fewer diagnostic errors and 13% fewer treatment errors. In absolute terms, the introduction of AI Consult would avert diagnostic errors in 22,000 visits and treatment errors in 29,000 visits annually at Penda alone. In a survey of clinicians with AI Consult, all clinicians said that AI Consult improved the quality of care they delivered, with 75% saying the effect was "substantial". These results required a clinical workflow-aligned AI Consult implementation and active deployment to encourage clinician uptake. We hope this study demonstrates the potential for LLM-based clinical decision support tools to reduce errors in real-world settings and provides a practical framework for advancing responsible adoption.

Yoshua Bengio, Tegan Maharaj, Luke Ong et al. · cmu, mila

Rapidly improving AI capabilities and autonomy hold significant promise of transformation, but are also driving vigorous debate on how to ensure that AI is safe, i.e., trustworthy, reliable, and secure. Building a trusted ecosystem is therefore essential -- it helps people embrace AI with confidence and gives maximal space for innovation while avoiding backlash. The "2025 Singapore Conference on AI (SCAI): International Scientific Exchange on AI Safety" aimed to support research in this space by bringing together AI scientists across geographies to identify and synthesise research priorities in AI safety. This resulting report builds on the International AI Safety Report chaired by Yoshua Bengio and backed by 33 governments. By adopting a defence-in-depth model, this report organises AI safety research domains into three types: challenges with creating trustworthy AI systems (Development), challenges with evaluating their risks (Assessment), and challenges with monitoring and intervening after deployment (Control).

Arvind Neelakantan, Tao Xu, Raul Puri et al.

Text embeddings are useful features in many applications such as semantic search and computing text similarity. Previous work typically trains models customized for different use cases, varying in dataset choice, training objective and model architecture. In this work, we show that contrastive pre-training on unsupervised data at scale leads to high quality vector representations of text and code. The same unsupervised text embeddings that achieve new state-of-the-art results in linear-probe classification also display impressive semantic search capabilities and sometimes even perform competitively with fine-tuned models. On linear-probe classification accuracy averaging over 7 tasks, our best unsupervised model achieves a relative improvement of 4% and 1.8% over previous best unsupervised and supervised text embedding models respectively. The same text embeddings when evaluated on large-scale semantic search attains a relative improvement of 23.4%, 14.7%, and 10.6% over previous best unsupervised methods on MSMARCO, Natural Questions and TriviaQA benchmarks, respectively. Similarly to text embeddings, we train code embedding models on (text, code) pairs, obtaining a 20.8% relative improvement over prior best work on code search.