Training in San Francisco

About the Team Training Runtime designs the core distributed machine-learning training runtime that powers everything from early research experiments to frontier-scale model runs. With a dual mandate to accelerate researchers and enable frontier scale, we’re building a unified, modular runtime that meets researchers where they are and moves with them up the scaling curve. Our work focuses on three pillars: high-performance, asynchronous, zero-copy tensor and optimizer-state-aware data movement; performant, high-uptime, fault-tolerant training frameworks (training loop, state management, resilient checkpointing, deterministic orchestration, and observability); and distributed process management for long-lived, job-specific and user-provided processes. We integrate proven large-scale capabilities into a composable, developer-facing runtime so teams can iterate quickly and run reliably at any scale, partnering closely with model-stack, research, and platform teams. Success for us is measured by raising both training throughput (how fast models train) and researcher throughput (how fast ideas become experiments and products). About the Role As a Training: ML Framework Engineer, you will work on improving the training throughput for our internal training framework, while enabling researchers to experiment with new ideas. This requires good engineering (for example designing, implementing, and optimizing state-of-the-art AI models), writing bug-free machine learning code (surprisingly difficult!), and acquiring deep knowledge of the performance of supercomputers. In all the projects this role pursues, the ultimate goal is to push the field forward. We’re looking for people who love optimizing performance, understanding distributed systems, and who cannot stand having bugs in their code. Since our training framework is used for large runs with massive numbers of GPUs, performance improvements here will have a large impact. This role is based in San Francisco, CA. We use a

About the Team Training Runtime designs the core distributed machine-learning training runtime that powers everything from early research experiments to frontier-scale model runs. With a dual mandate to accelerate researchers and enable frontier scale, we’re building a unified, modular runtime that meets researchers where they are and moves with them up the scaling curve. Our work focuses on three pillars: high-performance, asynchronous, zero-copy tensor and optimizer-state-aware data movement; performant, high-uptime, fault-tolerant training frameworks (training loop, state management, resilient checkpointing, deterministic orchestration, and observability); and distributed process management for long-lived, job-specific and user-provided processes. We integrate proven large-scale capabilities into a composable, developer-facing runtime so teams can iterate quickly and run reliably at any scale, partnering closely with model-stack, research, and platform teams. Success for us is measured by raising both training throughput (how fast models train) and researcher throughput (how fast ideas become experiments and products). About the Role As a Training Performance Engineer, you’ll drive efficiency improvements across our distributed training stack. You’ll analyze large-scale training runs, identify utilization gaps, and design optimizations that push the boundaries of throughput and uptime. This role blends deep systems understanding with practical performance engineering — analyzing GPU kernel performance, collective communication throughput, investigating I/O bottlenecks, and sharding our models so we can train them at massive scale. You’ll help ensure that our clusters are running at peak performance, enabling OpenAI to train larger, more capable models with the same compute budget. This role is based in San Francisco, CA. We use a hybrid work model of three days in the office per week and offer relocation assistance to new employees. In this role, you will: Profil

ABOUT BASETEN Baseten powers mission-critical inference for the world's most dynamic AI companies, like Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma and Writer. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, we enable companies operating at the frontier of AI to bring cutting-edge models into production. We're growing quickly and recently raised our $1.5B Series F , led by Altimeter Capital, Conviction Partners, and Spark Capital. Join us and help build the platform engineers turn to to ship AI products. We are looking for an engineer with strong experience in machine learning and solid foundations in maths and computer science to join our growing Post-Training team at Baseten. Custom models are instrumental to the success of Baseten customers. By inference volume, the overwhelming majority of traffic at Baseten is to and from models that have been post-trained in some way, whether that be through reinforcement learning, supervised finetuning, a recent technique from the literature, or an in-house research technique from Baseten. The Post-Training team is responsible for the success of our customers’ post-trained models, and we employ a wide array of techniques to produce models that are more efficient and higher quality than even the biggest closed source models for the customer’s specific needs. Your role as a research engineer is to build the in-house tooling to support all of this. We care about training a wide spectrum of different model architectures with a variety of techniques efficiently and at scale. At times this involves zooming deep into a particular technical topic, but more often if involves working across the stack as a whole - systems-level concepts like Kubernetes, cgroups, storage systems, and networking topologies, as well as PyTorch distributed tensor computation, and GPU kernels. RECENT RESEARCH Dense, on-policy or both? Repeated kv cache for long-running agents Distillation without the dark – rep

ABOUT BASETEN Baseten powers mission-critical inference for the world's most dynamic AI companies, like Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma and Writer. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, we enable companies operating at the frontier of AI to bring cutting-edge models into production. We're growing quickly and recently raised our $1.5B Series F , led by Altimeter Capital, Conviction Partners, and Spark Capital. Join us and help build the platform engineers turn to to ship AI products. THE ROLE This role sits at the frontier of our research agenda. You will pursue open problems at the intersection of post-training methodology and performant inference, and then collaborate with research engineering to translate findings into production systems. A meaningful portion of your time will be dedicated to research that deepens our understanding of how models learn, alignment, and architectural efficiency — questions that may not have immediate product application. The remainder will be directed toward research that solves concrete problems for Baseten's platform and customers, who are the fastest growing AI companies in the world like Cursor, Lovable, and Notion. We are looking for someone with sharp research taste and genuine creative instinct for problem selection. Someone who can identify questions that matter, design clean experiments to answer them, and push the state of the art. The environment here is not theoretical, but rather research that can be validated with eager customers who are serving billions of tokens a second. RECENT RESEARCH Towards infinite context windows: neural KV cache compaction Dense, on-policy or both? Repeated kv cache for long-running agents Distillation without the dark – replicating black-box on-policy distillation on Baseten RESPONSIBILITIES Define and pursue a research agenda spanning both foundational and applied work, with the applied component connected to Baseten's pla

ABOUT BASETEN Baseten powers mission-critical inference for the world's most dynamic AI companies, like Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma and Writer. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, we enable companies operating at the frontier of AI to bring cutting-edge models into production. We're growing quickly and recently raised our $1.5B Series F , led by Altimeter Capital, Conviction Partners, and Spark Capital. Join us and help build the platform engineers turn to to ship AI products. THE ROLE Forward Deployed Engineers work directly with the largest and fastest-growing AI companies in the world, owning their technical outcomes on Baseten and taking on the hardest problems in serving and improving models at scale. The work spans the model lifecycle: inference, post-training, and the systems that tighten the loop between them. Act as each account's de facto CTO on Baseten, with final accountability for how their workloads are designed, run, and scaled. Take customer objectives from vague to shipped: frame the problem, define the spec and success criteria, build the PoC, and carry it through to production quickly, using the right tools for the problem. Design the evals and benchmarks that isolate where quality or performance falls short, then close the gap yourself, whether that means optimizing inference, improving the model through post-training, or reworking the eval itself. Be the first responder to mission-critical failures including triage, owning the fix directly or route to the owning team and stay accountable until it ships. Build internal systems so that each engagement is faster than the last. This includes tooling and automation for eval and deployment infrastructure, and the recipes and reference implementations that make the product more self-serve. Shape the product itself, channeling what your accounts need into the roadmap and shipping fixes and features into Baseten's codebase yourse

ABOUT BASETEN Baseten powers mission-critical inference for the world's most dynamic AI companies, like Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma and Writer. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, we enable companies operating at the frontier of AI to bring cutting-edge models into production. We're growing quickly and recently raised our $1.5B Series F , led by Altimeter Capital, Conviction Partners, and Spark Capital. Join us and help build the platform engineers turn to to ship AI products. THE ROLE We’re looking for a customer-obsessed software engineer to come ship with us. You’ll own features like multi-node training and products like serverless reinforcement learning (RL) from conception to MVP (and from MVP to GA!). You’ll work through the stack, architecting solutions from API and UI down to our infrastructure layer. You’ll fine tune models yourself to develop an understanding of user workflows. You’ll work closely with research engineers leveraging state-of-the-art training techniques to build experiences that accelerate model development and solve for real pain points. If you’re excited to dive deep into the training, let’s talk! THE PRODUCT Take a look at what we’ve built so far: Overview of the product so far Training docs overview Story of the Training product Research we've done EXAMPLE INITIATIVES Checkpointing Pipeline: Our checkpointing pipeline starts with automated checkpointing, a feature that ensures that versions of models created during training are automatically backed up to the cloud. Users are able to then deploy checkpoints seamlessly into inference servers, providing point-and-click integrations into inference frameworks like vLLM and Baseten’s Inference Stack. This enables customers to quickly evaluate the performance of their checkpoints with real traffic. Multinode training: Multinode training enables customers to easily run training jobs across multiple compute nodes, enablin

ABOUT BASETEN Baseten powers mission-critical inference for the world's most dynamic AI companies, like Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma and Writer. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, we enable companies operating at the frontier of AI to bring cutting-edge models into production. We're growing quickly and recently raised our $1.5B Series F , led by Altimeter Capital, Conviction Partners, and Spark Capital. Join us and help build the platform engineers turn to to ship AI products. THE ROLE As a Software Engineer at on the Training Infrastructure team, you'll architect and lead development of our training platform, supporting top tier research engineers and model developers. You'll make key technical decisions for the infrastructure enabling developers to deploy, scale, and monitor their workloads with high performance and reliability. You’ll own scheduling, storage, networking, reliability, and observability of technical systems in the training stack EXAMPLE INITIATIVES Take a look at what we’ve built so far: Overview of the product so far Training docs overview Story of the Training product Research we've done RESPONSIBILITIES Design and architect scalable infrastructure systems for our ML training platform (e.g. scheduling, storage, and networking) Partner closely with developers and research engineers to translate complex training requirements into technical solutions Design and architect a global training scheduler Design and architect reinforcement learning systems and continuous learning pipelines Drive long-term improvements to improve reliability of systems and velocity of development Partner closely with SRE and Capacity teams to unlock state of the art training infrastructure Make critical architectural decisions balancing performance with system reliability Lead technical discussions and mentor junior engineers on infrastructure best practices Contribute to long-term technical strateg

About the Team The Agent Post-Training team creates the frontier agents OpenAI ships to the world. We are training the models behind our agents in Codex, ChatGPT, the API, and other frontier products: persistent, proactive intelligence that can operate computers, collaborate with people and other agents, and expand what people and organizations can imagine, attempt, and achieve. We define what the next generation of agents should be able to do, build the training signal that teaches those abilities, and run the experiments that make them real. Our work spans coding, tool use, computer use, multi-agent coordination, long-horizon execution, factuality, instruction following, calibrated reasoning, and taste. Our team is where new model capabilities get made. We build the data, environments, graders, training methods, and feedback loops that shape what OpenAI's next agents can do, then carry those capabilities through major training runs and into the products people use. About the Role As a member of Agent Post-Training, Connectors, you will teach models how to interface with the top professional software using code. You will help train agents to use code, APIs, tools, and structured integrations to operate across applications like Slack, Google Workspace, GitHub, Notion, Linear, Salesforce, and other core systems of work. You will help enable models to take useful actions across a user’s digital context: finding information, updating systems, coordinating work, generating artifacts, and completing multi-step workflows through the tools teams already use. You will train models to be supercharged by the world’s most important productivity and enterprise software, turning connected tools into a powerful action surface for our agents. You will work with researchers, engineers, product teams, infrastructure teams, and safety/alignment partners to decide what should go into major model runs, measure whether it worked, and ship improvements into products used by real people.

About the Team The Agent Post-Training team creates the frontier agents OpenAI ships to the world. We are training the models behind our agents in Codex, ChatGPT, the API, and other frontier products: persistent, proactive intelligence that can operate computers, collaborate with people and other agents, and expand what people and organizations can imagine, attempt, and achieve. We define what the next generation of agents should be able to do, build the training signal that teaches those abilities, and run the experiments that make them real. Our work spans coding, tool use, computer use, multi-agent coordination, long-horizon execution, factuality, instruction following, calibrated reasoning, and taste. Our team is where new model capabilities get made. We build the data, environments, graders, training methods, and feedback loops that shape what OpenAI's next agents can do, then carry those capabilities through major training runs and into the products people use. About the Role As a member of Agent Post-Training, you will improve the capabilities, reliability, and product fit of OpenAI's agentic models. You might own a research direction, build the infrastructure that makes large training runs faster and more trustworthy, create evals that reveal where models fail, or drive a capability from an idea through experimentation, integration, and launch. This role is intentionally broad. The strongest candidates are not defined by one method or subfield; they are people who can take an ambiguous capability problem and make progress across research, engineering, data, evals, and product. You should be excited to work on models that act in the world: writing and debugging code, using tools, calling functions, operating computers, collaborating with other agents, and completing valuable work on behalf of users. You will work with researchers, engineers, product teams, infrastructure teams, and safety/alignment partners to decide what should go into major model runs, meas

About the Team The Agent Post-Training team creates the frontier agents OpenAI ships to the world. We are training the models behind our agents in Codex, ChatGPT, the API, and other frontier products: persistent, proactive intelligence that can operate computers, collaborate with people and other agents, and expand what people and organizations can imagine, attempt, and achieve. We define what the next generation of agents should be able to do, build the training signal that teaches those abilities, and run the experiments that make them real. Our work spans coding, tool use, computer use, multi-agent coordination, long-horizon execution, factuality, instruction following, calibrated reasoning, and taste. Our team is where new model capabilities get made. We build the data, environments, graders, training methods, and feedback loops that shape what OpenAI's next agents can do, then carry those capabilities through major training runs and into the products people use. About the Role As a researcher working on Frontier Evals & Environments, you will help build north star model environments to drive progress towards safe AGI/ASI. Your work will directly guide the research programs of the most ambitious training runs happening at OpenAI. Some prior open-sourced evaluations built by researchers in this role include GDPval , SWE-bench Verified , MLE-bench , PaperBench , and SWE-Lancer . If you are interested in feeling firsthand the fast progress of our models, and steering them towards good outcomes, this is the role for you. You will work with researchers, engineers, product teams, infrastructure teams, and safety/alignment partners to decide what should go into major model runs, measure whether it worked, and ship improvements into products used by real people. This is a high-agency role for people who want their work to land directly in frontier models. In this role, you might Create ambitious RL environments to push our models to their limits, and measure frontie

About the Team The Agent Post-Training team creates the frontier agents OpenAI ships to the world. We are training the models behind our agents in Codex, ChatGPT, the API, and other frontier products: persistent, proactive intelligence that can operate computers, collaborate with people and other agents, and expand what people and organizations can imagine, attempt, and achieve. We define what the next generation of agents should be able to do, build the training signal that teaches those abilities, and run the experiments that make them real. Our work spans coding, tool use, computer use, multi-agent coordination, long-horizon execution, factuality, instruction following, calibrated reasoning, and taste. Our team builds the data, environments, graders, training methods, and feedback loops that shape what OpenAI’s next agents can do and what they are like to work with, then carries those improvements through major training runs and into products used by people every day. About the Role As a member of the Agent Post-training Personality team, you will help make OpenAI’s agents exceptional collaborators. You will study what makes an agent thoughtful, clear, perceptive, appropriately proactive, and genuinely easy to work with, then translate those insights into evals, training data, reward signals, and model improvements. We use “personality” to mean much more than writing style or general likability. It includes whether an agent understands what the user is trying to accomplish, communicates with good judgment, adapts to context, asks useful questions, handles disagreement honestly and takes initiative at the right moments. The goal is to create a strong, tasteful default that can adapt to different people and situations. This work combines behavioral research, product thinking, research and communication taste. You will collaborate with product teams, human experts, and researchers across post-training and pretraining to ensure that improvements survive the full trai

About the Team The Agent Post-Training team creates the frontier agents OpenAI ships to the world. We are training the models behind our agents in Codex, ChatGPT, the API, and other frontier products: persistent, proactive intelligence that can operate computers, collaborate with people and other agents, and expand what people and organizations can imagine, attempt, and achieve. We define what the next generation of agents should be able to do, build the training signal that teaches those abilities, and run the experiments that make them real. Our work spans coding, tool use, computer use, multi-agent coordination, long-horizon execution, factuality, instruction following, calibrated reasoning, and taste. Our team is where new model capabilities get made. We build the data, environments, graders, training methods, and feedback loops that shape what OpenAI's next agents can do, then carry those capabilities through major training runs and into the products people use. About the Role We believe that the final enabler for AGI is spending compute on context. As a Context Researcher on Agent Post-Training, you will scale compute spent on context. You will get to work in our frontier training stack on enabling the next paradigm of model training with a clear product interface for iterative deployment (Codex Chronicle). You will work with researchers, engineers, product teams, infrastructure teams, and safety/alignment partners to decide what should go into major model runs, measure whether it worked, and ship improvements into products used by real people. This is a high-agency role for people who want their work to land directly in frontier models. In this role, you might Design and run experiments that improve scaling of compute on context. Own end-to-end improvements to the post-training stack, including RL, data pipelines, graders, reward signals, evals, diagnostics, and model-behavior analysis. Build evals and environments that expose the next set of model failures,

About the Team Training Runtime builds the distributed systems that power OpenAI's largest model training runs - most recently GPT-5.5! The Data Movement area owns the infrastructure that keeps training jobs supplied with the right data at the right time, and keeps model state moving safely and efficiently across large clusters. Our work spans machine learning systems, distributed storage, high-throughput data loading, reliability engineering, and developer experience. Success means researchers can move quickly while training runs remain fast, reproducible, debuggable, and resilient at scale. About the Role We are looking for a deeply hands-on Technical Lead Manager to own datasets throughout our training infrastructure. This person will set the direction for how training jobs read data: the APIs, storage contracts, versioning model, benchmarks, debugging tools, and reliability guarantees that make data access consistent across current and future training frameworks. You will begin as the primary technical owner for dataset reads, working directly in the code while aligning researchers, training framework owners, storage teams, and infrastructure partners around a durable platform. The problem is deceptively hard at frontier scale: make enormous, heterogeneous datasets easy to consume, correct across distributed workers, observable when something goes wrong, and flexible enough to support pretraining, reinforcement learning, and multimodal training. In this role, you will Design and build a unified dataset read platform for multiple current and future training frameworks. Define dataset APIs, storage-format expectations, registration/versioning, and migration paths that make data access reproducible and maintainable. Build reliability into the read path, including stateful iteration, caching, fast restart, recovery, and clear operational contracts. Build terminal and web-based visualizers that let teams inspect text, multimodal, and reinforcement learning data late

About the Team The Agent Post-Training team creates the frontier agents OpenAI ships to the world. We are training the models behind our agents in Codex, ChatGPT, the API, and other frontier products: persistent, proactive intelligence that can operate computers, collaborate with people and other agents, and expand what people and organizations can imagine, attempt, and achieve. We define what the next generation of agents should be able to do, build the training signal that teaches those abilities, and run the experiments that make them real. Our work spans coding, tool use, computer use, multi-agent coordination, long-horizon execution, factuality, instruction following, calibrated reasoning, and taste. Our team is where new model capabilities get made. We build the data, environments, graders, training methods, and feedback loops that shape what OpenAI's next agents can do, then carry those capabilities through major training runs and into the products people use. About the Role As a member of this API & power-users team, you will improve the capabilities, reliability, and product fit of OpenAI’s agentic models for power users and API developers. You might design evals from real developer workflows, build training environments around production-like tool use, turn qualitative model failures into training data, evals, or post-training interventions, or drive a behavior improvement from discovery through post-training, integration, and launch. This role is intentionally broad. The strongest candidates are comfortable turning ambiguous model behavior problems into concrete progress, whether that means improving tool use, planning, instruction following, recovery from mistakes, or how models behave in API-based workflows. You should be excited to work across research, engineering, data, evals, and product to make models better at acting in real workflows. You will work closely with researchers, engineers, API/product teams, Codex, infrastructure, and safety/align

About the Team The Agent Post-Training team creates the frontier agents OpenAI ships to the world. We are training the models behind our agents in Codex, ChatGPT, the API, and other frontier products: persistent, proactive intelligence that can operate computers, collaborate with people and other agents, and expand what people and organizations can imagine, attempt, and achieve. We define what the next generation of agents should be able to do, build the training signal that teaches those abilities, and run the experiments that make them real. Our work spans coding, tool use, computer use, multi-agent coordination, long-horizon execution, factuality, instruction following, calibrated reasoning, and taste. Our team is where new model capabilities get made. We build the data, environments, graders, training methods, and feedback loops that shape what OpenAI's next agents can do, then carry those capabilities through major training runs and into the products people use. About the Role As a member of Agent Post-Training, Computer Use, you will teach models to operate computers. You will help train models that can navigate browsers and desktops, use tools and applications, reason through complex workflows, collaborate with users and other agents, and complete long-horizon tasks with reliability and judgment. This work sits at the intersection of frontier model training, product behavior, evaluation, and systems engineering, and will directly shape the computer-use capabilities shipped in OpenAI’s next generation of agents. Currently, our models are the best in the world at this behavior! You will work with researchers, engineers, product teams, infrastructure teams, and safety/alignment partners to decide what should go into major model runs, measure whether it worked, and ship improvements into products used by real people. This is a high-agency role for people who want their work to land directly in frontier models. In this role, you might Design and run experiments th