Google DeepMind is hiring for Polaris.
Google DeepMind is hiring engineers for Polaris, a new team that builds tests to measure how well Gemini handles real software engineering work. The pay stands out. Total compensation runs from $550,000 to $1.1 million a year, and you don’t need a degree to apply.Tamay Besiroglu, an Epoch AI co-founder who now works on data at Google DeepMind, kept the brief simple in a post on X. “All we care about is you being smart and able to solve hard problems,” he wrote, adding that people should ignore the pay mentioned in the job ad. When asked whether the listing was wrong, he clarified that its numbers were accurate but left out equity. The base salary alone sits between $164,000 and $252,000.
Every Polaris test for Gemini takes about a week to build
The role, listed as Research Engineer and based in Mountain View, California, centres on designing tasks. Each one is a detailed exam for Gemini, made up of a prompt telling the model what to do and a grader or rubric that scores its attempt.Instead of producing tests in bulk through a synthetic data pipeline, Polaris engineers hand-build a smaller set that captures the complexity of real-world coding. A single task usually takes about a week, from first idea to final approval.
Engineers will also start new projects to improve Gemini’s training data and evaluations. Google first called for Polaris applicants on September 25, describing it as a new DeepMind team designing evaluation frameworks for AI coding agents.
A very good vibe coder can land the Google DeepMind job
On paper, the minimum bar asks for software development experience and hands-on use of AI coding tools and developer agents. A track record at contests like the IMO, Putnam or IOI, or experience designing benchmarks and evaluation rubrics, counts in your favour.Besiroglu’s replies on X widened the door further. Asked if non-engineers could apply, he said quick learners were fine. Asked about vibe coders, he said they just need to be very good ones. He also confirmed the team doesn’t ask LeetCode questions. When one user called Google’s hiring process “laughably bad”, Besiroglu said his own team runs the interviews and that he thinks the process is good.So what is DeepMind looking for in an engineer? Someone smart who can solve hard problems, Besiroglu said. Experience with AI coding agents, he added, is very useful too.