Anthropic OA Questions 2026: What Candidates Actually Saw
Last updated: June 2026
5+ Anthropic online assessment questions reported by candidates who took the OA in 2026. Sourced from 1Point3Acres, LeetCode, Blind, Glassdoor, and Reddit. Data grows as new reports are added.
Anthropic OA Format in 2026
Anthropic uses CodeSignal for its online assessment. Candidates report a multi-level problem structure where each level unlocks on completion of the prior one. The most frequently cited problem type is a Memory Database design challenge requiring layered implementation of get, set, and delete operations with TTL and snapshot support. The assessment tests both coding speed and API-design judgment under time pressure.
Anthropic OA Topic Patterns
Reported topics skew heavily toward data structure design (hash maps, queues, stacks), string parsing, and multi-step implementation problems. Pure algorithmic puzzles (tree traversal, dynamic programming) appear less often than design-style problems that test clean code organization.
Real Anthropic OA Questions from 2026
Anthropic Software Engineer Interview OA and Phone Screen
I noticed not many people have shared their experience interviewing with Anthropic. I'm currently going through the process and just went through the phone screen, so I figured I'd share and help out.
anthropic fellows two-round codesignal online assessment review for machine learning intern
分享一下 人类学Fellows 的两轮 CodeSignal 面经，希望对后面申请的人有帮助 四月中申请的前几天收到， 第一关做完后马上收到第二关于是接着给他做完了 具体time line 如下 然后都是proctored开摄像头麦克风 以下内容需要积分高于 188 您已经可以浏览 第一轮是 90 分钟的 general coding assessment，我遇到的是一个偏系统/网络模拟的 Python 题，大方向是实现一个简化版的 DNS resolver。题目会分成多个 step，每一步对应一组 unit tests，需要逐步实现功能。整体不是传统 LeetCode 那种纯算法题，更像是给你一个 mini codebase 和清晰 spec，然后要求你按测试一步步补全逻辑。 难度体感上，我觉得第一轮大概是 LeetCode Medium 左右，但不是算法难，而是工程逻辑和边界处理比较多。前面几步比较 straightforward，比如字符串 normalize、基本递归/迭代查询；后面会逐渐加入更多情况，例如 alias / fallback / error hand...
anthropic online assessment and interview timeline for software engineer role
Recipe Manager 3号做的oa 600/600 今天才收到邮件schedule recruiter call。一度以为默拒了。 QQ：感觉现在面的组和background并没有特别match，和recruiter说换个组面可以吗？
anthropic online assessment experience for software engineer fulltime role
以下内容需要积分高于 200 您已经可以浏览 我的prompt是The project will be a toy simulation of an app that won't require anything except the standard library of your chosen programming language。一开始我还以为是in-memory DB，但实际上最后是cloud storage。我觉得现在OA可能变成了统一的prompt，然后题目是随机的。 题目跟hack2hire上面一模一样。本身没有什么复杂的算法，但如果level 1时候的OOP还有数据结构没有设计好，后面可能就需要refactor，会浪费很多时间
anthropic online assessment: debugging extremely randomized tree with numpy
本帖最后由 小鬼佳佳 于 2026-5-30 21:38 编辑 numpy debug 是实现一个extremely randomized tree 的numpy debug. 难度比第一轮大。第一轮是dns solver. 求加米看 take home。 回来再分享take home 题。 求一个5h takehome。
Topic Breakdown (2026 OA)
Topics reported by candidates who took the Anthropic OA in 2026:
How to Prepare for the Anthropic OA
Anthropic OA prep should focus on multi-level implementation problems rather than pure algorithm drilling. Practice building small in-memory systems (caches, queues, event logs) under timed conditions. CodeSignal uses a proprietary score that penalizes test failures and time taken, so test coverage and edge case handling are as important as speed. Review string manipulation and hash map patterns thoroughly. Candidates who have taken the Anthropic OA consistently recommend practicing on CodeSignal's General Coding Framework before the actual assessment.
The Anthropic OA is a timed, auto-graded coding test. Preparation is different from interview prep: you need to optimize for both correctness and speed, since partial credit requires passing test cases, not just showing reasoning.
- The assessment runs on CodeSignal, with a 90 to 120 minutes window. Familiarize yourself with the platform's editor and I/O format before test day.
- Practice timed sessions. Set a timer for 35 to 45 minutes per problem and simulate the real constraint.
- Topic emphasis: see the Topic Patterns section above for Anthropic-specific prep signals.
- Read the problem constraints carefully before writing code. The constraints often reveal the expected time complexity and algorithm type.
- Use the LeakCode database to review Anthropic OA questions from actual candidates. Pattern recognition across multiple sightings is a stronger prep signal than any single mock problem.
- If the OA includes a work simulation section, practice under realistic conditions. This section tests judgment and values alignment, not coding ability.
FAQ
What does the Anthropic OA look like in 2026?
Anthropic uses CodeSignal for its online assessment. Candidates report a multi-level problem structure where each level unlocks on completion of the prior one. The most frequently cited problem type is a Memory Database design challenge requiring layered implementation of get, set, and delete operations with TTL and snapshot support. The assessment tests both coding speed and API-design judgment under time pressure.
How long is the Anthropic OA?
The Anthropic OA runs 90 to 120 minutes on CodeSignal. Exact timing for your role may vary. Check your invitation email for the specific window allocated.
What platform does Anthropic use for the OA?
Anthropic uses CodeSignal for its online assessment. The platform supports multiple languages including Python, Java, C++, and JavaScript. Your invitation email will include the access link and any platform-specific instructions.
Are these real Anthropic OA questions?
Yes. All questions in the LeakCode database come from candidate reports on 1Point3Acres, Blind, Glassdoor, Reddit, and LeetCode. They are not AI-generated. Each question links to its source where available.
Other Anthropic OA Years
Related: Anthropic OA Hub · Anthropic All Questions · All OA Questions · Data Sources