# QuizPilot Canonical page: https://quizpilot.link/en/ QuizPilot is a Chrome extension: an AI study helper for practice questions on the web. On a practice, homework or self-test page it reads the questions, suggests an answer with a confidence score for each, and then ticks or fills them in (one-click undo, never submits) or only marks them on the page. ## What it does - Question types: single choice, multiple answer, true/false, fill-in, short answer, and questions with pictures. When the page text is incomplete or a question is an image or canvas, a screenshot goes to a multimodal model. - Sites: not tied to particular websites. It infers questions, options and answer controls from the page structure, so most online practice, question bank and quiz pages work as they are. An unusual layout can be learned once from a screenshot. - Ways to run it: "Solve this page" (fills answers, one-click undo), "Assist" (only marks the suggested answer), "Continuous" (one-question-per-page quizzes: answers and moves on, stops at the last question), select an area, right-click menu, keyboard shortcuts (Alt+Shift+1–4; Control+Shift on Mac). - Languages: English, Chinese, Japanese and Korean interface. ## Free mode and Paid mode - Free mode: bring your own model API key (Jev via OpenRouter or TypeSafe, or any OpenAI-compatible service, local models included). The key stays in the browser; questions go straight to that provider, never through QuizPilot's servers. - Paid mode: sign in with email or Google and pay per answered question. Two answer modes, switchable in the popup: - Fast (each answered question counts as 1): multiple-choice and true/false questions go to TypeSafe's Jev decision model; fill-in, short answer and picture questions to a multimodal model. Optional low-confidence review re-checks unsure answers. - Accurate (each answered question counts as 10): every question goes to an advanced AI model that reasons before answering. Slower, more accurate on hard multi-step questions. - Details: [Pricing](https://quizpilot.link/pricing.md) ## Measured accuracy On the developer's own 30 multi-step reasoning questions, Fast got 26–27 right and Accurate all 30. On 60 basic questions Jev got 59. Every true/false and multiple-select answer from Jev was right; its misses were single-choice questions needing multi-step arithmetic. Full write-up: [How we use Jev](https://quizpilot.link/blog/jev.md). ## Privacy The extension reads the current page only when the user clicks its button, uses the right-click menu or presses a shortcut, never in the background. Paid mode passes questions through QuizPilot's server to the model and does not keep questions, screenshots or answers. Policy: https://quizpilot.link/en/privacy ## Requirements and install - Desktop Google Chrome 116 or later on Windows, macOS or Linux. - Install: [Chrome Web Store](https://chromewebstore.google.com/detail/quizpilot/gloldlbaimjdhcpgpcdcihcekfclgeha) ## Open source The extension and model adapters are open source under the AGPL-3.0: https://github.com/jelly-ham/quizpilot-extension. The Paid mode backend is not. ## Responsible use A study aid for practice, homework and self-tests where such tools are allowed. Not for exams or assessments that prohibit them; it never submits answers. ## More - [Pricing](https://quizpilot.link/pricing.md) - [FAQ](https://quizpilot.link/faq.md) - [中文介绍](https://quizpilot.link/zh.md) - [How we use Jev (benchmark)](https://quizpilot.link/blog/jev.md) - Contact: support@quizpilot.link --- # QuizPilot pricing Canonical page: https://quizpilot.link/en/#pricing ## Free mode Free. Add your own model API key; the provider you choose bills you for model use. ## Paid mode Balance is counted in questions. Only answered questions are charged; a question the model could not answer costs nothing. Questions do not expire. | Mode | Each answered question counts as | Best for | |---|---|---| | Fast | 1 question | everyday practice; answers in a few seconds | | Accurate | 10 questions | hard multi-step problems; a little slower | Reading a page from a screenshot counts as 1 question. | Pack | Price (USD) | Fast answers | Accurate answers | |---|---|---|---| | Starter | $3 | 450 | 45 | | | $5 | 1,000 | 100 | | | $10 | 4,000 | 400 | | | $20 | 11,000 | 1,100 | New accounts get 200 free questions. Payment by card through Stripe. Prices as of October 2026; the live page is authoritative. Install: [Chrome Web Store](https://chromewebstore.google.com/detail/quizpilot/gloldlbaimjdhcpgpcdcihcekfclgeha) --- # QuizPilot FAQ Canonical page: https://quizpilot.link/en/#faq ## What is QuizPilot? QuizPilot is a Chrome extension: an AI study helper for practice questions on the web. One click on a practice, homework or self-test page reads the questions and suggests answers with a confidence score. It can tick or fill in the answers, or only mark them on the page. ## Is QuizPilot free? Free mode costs nothing: add your own model API key, and the provider you chose bills you for model use. Paid mode works right after sign-in and charges per answered question: 1 question in Fast mode, 10 in Accurate mode, nothing for questions without an answer. New accounts get 200 free questions, and packs start at $3 for 450 questions. ## Which question types does it handle? Multiple choice, multiple answer, true/false, fill-in, short answer, and questions with pictures. When the page text is incomplete or the question is an image or canvas, a screenshot goes to a multimodal model. ## Which websites does it work on? It is not tied to particular sites: it infers questions, options and answer controls from the page structure, so most online practice, question bank and quiz pages work as they are. For an unusual site, "Relearn this page's layout" learns it once; if it still reads the page wrong, you can send us the debug log from Settings. ## Is QuizPilot an AI quiz solver? Yes. QuizPilot is an AI quiz solver and homework helper for practice questions on the web: it reads a question and returns an answer with a confidence score, then fills it in or only marks it. It works on most online practice, question bank and quiz pages without site-specific setup. ## Is QuizPilot a homework helper Chrome extension? Yes — it installs from the Chrome Web Store and runs on desktop Chrome 116 or later. As a homework helper it reads practice and homework questions, suggests answers with a confidence score, and never submits for you. ## Will it submit my answers? No. QuizPilot only ticks or fills in answers, and you can undo that in one click; submitting is always up to you. Continuous mode stops at the last question. ## Is my data safe? The extension reads the current page only when you click its button, use the right-click menu or press the shortcut, never in the background. In Free mode your key stays in the browser and questions go straight to the provider you chose. In Paid mode questions pass through our server to the model, and we do not keep the questions, screenshots or answers. See the Privacy Policy. ## Can I use it in exams? Not in exams or assessments that prohibit such tools. QuizPilot is a study aid: use it only for practice, homework and self-tests where it is allowed, and follow the rules of your school, exam body and the website. ## Which AI models does it use? Paid mode has two modes. Fast: multiple-choice and true/false questions go to TypeSafe's Jev decision model, and fill-in, short answer and picture questions to a multimodal model; turn on low-confidence review in Settings to have unsure answers checked by a stronger model. Accurate: every question goes to an advanced AI model. Free mode uses the models you set up. ## What is the difference between Fast and Accurate? Fast answers with Jev and a light LLM in a few seconds and counts each question as 1: good for everyday practice. Accurate sends every question to an advanced AI model that thinks before it answers: more accurate on hard questions, a little slower, and each question counts as 10. On our own set of 30 multi-step reasoning questions, Fast got 26–27 right and Accurate all 30. In both modes questions without an answer are free, and you can switch any time in the extension popup. ## Which browsers and systems are supported? Desktop Chrome 116 or later on Windows, macOS and Linux. Install it free from the Chrome Web Store. ## Is QuizPilot open source? The extension and its model adapters are open source on GitHub under the AGPL-3.0 (github.com/jelly-ham/quizpilot-extension), so the Free mode privacy claims can be checked in the code. The Paid mode backend is not open source. --- # How we use Jev to answer quiz questions (with a 90-question benchmark) Canonical page: https://quizpilot.link/en/blog/jev/ QuizPilot is an open-source Chrome extension that reads practice questions on a web page and suggests answers. Since TypeSafe released Jev, our Fast mode sends every multiple-choice and true/false question to it. This post covers how we map quiz questions onto Jev, and what a small benchmark showed. ## The short version - **Accuracy:** Jev answered 85–86 of 90 questions correctly across two runs. A lightweight LLM got 89, a frontier reasoning LLM got all 90. - **Speed:** one Jev request answered all 60 basic questions in 0.17–0.44 s. The lightweight LLM took 2.6 s, the reasoning LLM 8.5 s. - **Where it slips:** every true/false and multiple-select answer was right. All of its mistakes were single-choice questions that need several steps of arithmetic. - **Confidence is useful:** 7 of Jev's 9 wrong answers came back with a confidence below 0.6, against an average of 0.93 for its right answers. ## Why a decision model for quiz questions A multiple-choice question already lists every possible answer. An LLM still writes its reply as text, and we parse the letter back out. Jev skips that: you give it a state and typed questions (pick one of these options, or yes/no), and it returns the choice with a probability. No prose to parse, no answer outside the options. ## How a quiz question becomes a Jev question - **Single choice** → one Jev `choice` question whose criteria are the options. - **True/false** → one yes/no (`noul`) question: "Is this statement true?" - **Multiple select** → one yes/no question per option: "Is option C one of the correct answers?" Every option at 0.5 or above is ticked; if none is, the likeliest one is. - **Fill-in, short answer, picture questions** → an LLM. Jev does not write text or read images. A trimmed request for one true/false question: ```json { "model": "typesafe/jev-1.13", "state": "Questions from a quiz or exercise on a web page. Each question is independent. Judge by factual correctness, not by wording or position of the options.", "questions": { "q0": { "type": "noul", "instructions": { "statement": "The harmonic series 1 + 1/2 + 1/3 + … converges.", "task": "Is this statement true?" }, "criteria": { "true": "The statement is correct", "false": "The statement is incorrect" } } } } ``` A yes/no answer is a single probability, so we turn it into a confidence by its distance from 0.5: `|2p − 1|`. A multiple-select question takes the lowest confidence among its options. A whole page goes out in one request, split only when it nears Jev's input limit. The adapter is open source: [github.com/jelly-ham/quizpilot-extension](https://github.com/jelly-ham/quizpilot-extension). ## The benchmark 90 questions with known answers, half English and half Chinese, none from a real exam: - **Basic set (60):** 20 single choice, 20 multiple select, 20 true/false; half easy, half harder. - **Hard set (30):** mostly multi-step problems, such as the trailing zeros of 100!, 2^100 mod 7, and inclusion–exclusion counting. Comparison models: a lightweight general LLM with thinking off, and a frontier reasoning LLM at medium effort. Jev and the lightweight LLM ran twice. | Model | Basic (60) | Hard (30) | Time, basic set | |---|---|---|---| | Jev | 59, 59 | 26, 27 | 0.17–0.44 s | | Lightweight LLM | 60 | 29, 29 | 2.6 s | | Reasoning LLM | 60 | 30 | 8.5 s | Jev by question type, both runs and both sets: true/false 60/60, multiple select 50/50, single choice 61/70. ## Where Jev gets it wrong All nine mistakes were single-choice questions whose answer has to be computed, not recognised: trailing zeros of 100!, 2^100 mod 7, a divisibility count from 1 to 100, the number of real roots of x³ − 3x = 1, and (once) the interior angles of a 12-gon. Fact-checking questions, including harder ones like "every group of prime order is cyclic", were all right. ## Confidence tells you when to ask again Across 180 Jev answers, 9 were wrong. Seven had a confidence between 0.26 and 0.49; the two exceptions were the same question (trailing zeros of 100!) at 0.69 and 0.71. Right answers averaged 0.93. Re-asking everything below 0.6 with the lightweight LLM would have touched 14 of 180 answers (8%). Using the LLM's answers from the same benchmark, the combination scores 60/60 on the basic set and 28/30 on the hard set. (Worked out from the separate runs, not a live run with review on.) ## Limits - 90 questions we wrote ourselves show large gaps, not small ones. - Text only; picture questions never go to Jev. - Jev 1.13 via OpenRouter, October 6, 2026. Full post with more detail: [quizpilot.link/en/blog/jev](https://quizpilot.link/en/blog/jev/). Happy to answer questions about the integration in the comments. --- # QuizPilot（中文介绍） 页面：https://quizpilot.link/ QuizPilot 是一款 Chrome 浏览器扩展，是面向网页练习题的 AI 学习助手：在练习、作业或自测页面上点一下，读取题目，给出参考答案和置信度，可以自动勾选或填写（可一键撤销，绝不自动提交），也可以只标注建议答案。 - 免费版：填入自己的模型 API Key，Key 只存在浏览器里，题目直接发给你选的模型服务商。 - 付费版：邮箱或 Google 登录，按答出的题数扣费。快速模式每题按 1 题计（选择题、判断题用 TypeSafe 的 Jev 决策模型）；精准模式每题按 10 题计（所有题交给高级大模型）。新账户送 200 题；充值 $3（450 题）、$5（1,000 题）、$10（4,000 题）、$20（11,000 题）。 - 安装：[Chrome 应用商店](https://chromewebstore.google.com/detail/quizpilot/gloldlbaimjdhcpgpcdcihcekfclgeha) - 开源：https://github.com/jelly-ham/quizpilot-extension（AGPL-3.0） ## 常见问题 ### QuizPilot 是什么？ QuizPilot 是一款 Chrome 浏览器扩展，是面向网页练习题的 AI 学习助手。在练习、作业或自测页面上点一下，它会读取题目，给出参考答案和置信度；可以自动勾选或填写，也可以只在页面上标出建议答案。 ### QuizPilot 免费吗？ 免费版不收费：填入你自己的模型 API Key 就能使用，模型费用由你选择的服务商收取。付费版登录即用，按答出的题数扣费：快速模式每题按 1 题计，精准模式每题按 10 题计，没答出来的题不扣。新账户赠送 200 题，充值 $3 起（450 题）。 ### 支持哪些题型？ 单选、多选、判断、填空、简答，以及需要看图的题目。页面文字读不全，或者题目是图片、画布时，会截图交给多模态模型识别。 ### 支持哪些网站？ 不针对特定网站：扩展按页面结构识别题目、选项和作答控件，常见的在线练习、题库和测验页面一般可以直接使用。遇到结构特殊的网站，可以用“重新识别本页结构”学习一次；仍然读不对的，可以在设置里把调试日志发给我们。 ### 会替我提交答卷吗？ 不会。QuizPilot 只勾选或填写答案，并且可以一键撤销；提交始终由你自己完成。连续答题到最后一题会停下。 ### 我的数据安全吗？ 扩展只在你点击按钮、使用右键菜单或按快捷键时读取当前页面，不会在后台读取你浏览的网页。免费版的 Key 只保存在浏览器里，题目直接发给你选择的模型服务商；付费版的题目经我们的服务器转发给模型，我们不保存题目、截图和答案。详见隐私政策。 ### 可以在考试中使用吗？ 不可以在禁止使用辅助工具的考试或测评中使用。QuizPilot 是学习辅助工具，请只在允许的练习、作业和自测中使用，并遵守学校、考试机构和网站的规定。 ### 用的是什么 AI 模型？ 付费版有两种模式。快速模式：选择题和判断题交给 TypeSafe 的 Jev 决策模型，填空、简答和看图题交给多模态大模型；在设置里开启“低置信度复核”后，把握不大的题会再交给更强的模型。精准模式：所有题目都交给高级大模型。免费版使用你自己配置的模型。 ### 快速模式和精准模式有什么区别？ 快速模式用 Jev 和轻量大模型作答，几秒出结果，每题按 1 题计，适合日常练习。精准模式把每道题交给高级大模型，先思考再作答，难题更准，但稍慢，每题按 10 题计。在我们自测的 30 道多步推理题里，快速模式答对 26–27 道，精准模式全部答对。两种模式都是没答出来的题不扣，可以在扩展弹窗里随时切换。 ### 支持哪些浏览器和系统？ 电脑版 Chrome 116 及以上，Windows、macOS 和 Linux 都可以使用。在 Chrome 应用商店免费安装。 ### QuizPilot 开源吗？ 扩展和模型适配层以 AGPL-3.0 许可证在 GitHub 开源（github.com/jelly-ham/quizpilot-extension），免费版的隐私承诺可以在代码里核实。付费版的服务端不开源。