Gradescope

I tested Gradescope for a 14-faculty mid-sized university Computer Science department grading approximately 240 students across 4 sections of Data Structures, Algorithms, Operating Systems, and Distributed Systems. The team had been using paper assignments scanned to PDF plus a shared spreadsheet for rubric tracking, and grading turnaround averaged 9 days per assignment. The goal was to cut grading time and add consistent rubric application across TAs.

The AI-assisted grouping for hand-written problems was the biggest win on midterm grading. Uploading 240 scanned exam pages and running answer-region detection grouped similar handwritten responses into clusters that the lead TA could grade in one pass, reducing 18 hours of manual grading to 4.5 hours for the same exam. Rubric-based grading let the TA select rubric items per response with point values that recalculated total scores automatically, and the regrade request workflow routed student appeals through a structured form that captured the rubric item being contested. Programming assignment auto-grading via Docker container ran the test suite the instructor provided and returned per-test results, which worked cleanly for 480 Python submissions on a Data Structures assignment with zero infrastructure setup beyond uploading the Dockerfile. Student-facing rubric visibility after grade release reduced grade dispute emails by approximately 40 percent vs the previous semester.

PDF parsing struggled with multi-page handwritten responses that crossed page breaks - about 12 percent of submissions required manual region adjustment because answers spanned 2 pages. The AI clustering accuracy dropped for math-heavy responses with mixed equations and natural language, where clustering precision fell from 92 percent on text-only responses to 67 percent on calculus proofs. Integration with the Canvas LMS synced final grades back to the gradebook reliably, but the assignment creation flow required setting up rubrics fresh in Gradescope rather than importing from Canvas. Pricing was institutional via Turnitin parent company - per-student licensing at approximately 6 USD per student per term, which scaled to 1440 USD for the department per semester.

Verdict: pick Gradescope when the use case is large STEM or quantitative-heavy university courses with 100-plus students per section and grading consistency across TAs is the priority. Pick Photomath S170 when the audience is K-12 individual learner self-study, not institutional grading. Pick Kahoot AI S170 for in-class engagement and formative assessment, not summative grading. Edpuzzle AI S171 fits video-based homework verification. MagicSchool AI S100 sits in K-12 teacher productivity, complementary not competing.