SubSchool - Create courses and grade open-ended answers with GPT-5.6

SubSchool uses GPT-5.6 to turn descriptions, books and video libraries into editable courses, and recorded tutoring sessions into follow-up homework. Terra powers everyday generation and grading; teachers can switch to Sol for complex tasks. Grade essays, interviews, photos, and presentations automatically based on grading criteria, keep teachers in control, and let Codex create or update real course content through MCP.

Add a comment

Replies

Best
We launched SubSchool on Product Hunt in March. Since then, we have rebuilt the teacher and student applications, moved the main generation and assessment workflows to GPT-5.6, and added a new way to operate the platform through Codex and MCP. SubSchool is built for teachers who already have knowledge and materials but do not want to spend days converting them into a course. A teacher can start with a description, a book or a set of video lectures. SubSchool creates editable modules, lessons, descriptions, exercises, and homework directly inside the platform. For one-on-one lessons, session recordings are transcribed to generate assignments based on what was actually discussed, rather than simply on the topic or title of the lesson. The same workflow continues after a student submits their work. SubSchool can automatically grade: - essays and other extended written answers; - spoken interviews; - photos of handwritten or visual solutions; - presentations. The teacher defines the assessment criteria and their weights. GPT-5.6 returns a score and criterion-specific feedback. The teacher can then review the original submission, change the grade, and add a final comment. Terra is the default model for everyday generation and grading. Teachers can switch to Sol in their settings for long source materials, curriculum design, complex exercise generation, or assessments that require more reasoning. We decided not to hide this behind automatic model routing. The teacher chooses the balance between cost and model capability. We also built a SubSchool MCP server. Codex can inspect, create, and update real platform entities such as courses, lessons, and exercises through structured tools. The result is saved directly in SubSchool instead of being generated in one tool and copied manually into another. GPT-5.6 was also part of how we built this release. We used Sol in Codex Ultra mode to work across the existing codebase, develop the MCP integration and Events section, and ship the redesigned teacher and student web applications. The release also includes: - a new student mobile app; - Events; - tutor discovery and lesson requests; - redesigned teacher and student web applications. The current applications are available at: teacher.subschool.us class.subschool.us The previous versions remain available at: teacher-old.subschool.us class-old.subschool.us The old applications are still online, so the redesign and workflow changes can be compared directly rather than judged from selected screenshots. The part we are still working through is where automation should stop. Which actions would you allow to run automatically after a teacher has configured them, and which should always require explicit teacher approval?

What the AI ​​generates effectively automates the process. But do you have specialists for quality control?