Why UC Berkeley CS Students Are Failing at Record Rates

UC Berkeley's computer science program recorded its worst grades in recent memory during spring 2026. In CS 10, an introductory course, 35.3% of students received failing grades. In CS 61A, the foundational programming course for majors, 10.6% failed. According to The Daily Californian, both figures represent a sharp jump from the previous two years, when failure rates in these courses stayed below 10%.

These numbers also fall well short of the EECS department's targets. The department expects no more than 7% of lower-division students to receive D's and F's combined. Both CS 10 and CS 61A posted average grades of C-plus—equivalent to a 2.3 GPA—well below the department's goal of 2.8 to 3.3.

The Problem Reaches Beyond Introductory Courses

The grade decline is not limited to entry-level classes. EECS 127, an upper-level optimization course taught by associate teaching professor Gireeja Ranade, recorded a 16.8% failure rate—more than triple the department's typical 5% benchmark for D's and F's in advanced courses.

Professor Dan Garcia, who taught both CS 10 and CS 61A in spring 2026, reported that nearly 30 students in CS 10 were caught cheating on take-home exams. Garcia designed the Beauty and Joy of Computing (BJC) curriculum, which forms the backbone of CS 10 and Berkeley's pre-college summer computer science academy.

What's Driving the Decline

Faculty members point to two main culprits: increased use of AI tools like ChatGPT, and students arriving without strong math skills. The concern is widespread—more than 1,300 UC faculty members have signed a petition raising these academic issues.

A Berkeley study found roughly a 30% jump in A grades after ChatGPT's release. This suggests AI tools initially inflated grades before the university developed better detection systems and changed how it assesses student work. Faculty also cite staffing shortages as a contributing factor.

This pattern echoes something we have seen before. When programmable calculators arrived in the 1980s, internet search engines in the 1990s, and smartphone apps in the 2000s, each sparked a similar period where grading initially became unreliable before educators figured out how to assess skills in the new environment. This time, the shift to AI appears to be happening faster than those previous transitions.

What This Means for the Pipeline

These failure rates raise a real question: what happens to the pipeline of computer science talent. CS 10 brings in students who are curious about computational thinking but may not be CS majors. CS 61A is the entry point for students who intend to major in computer science. High failure rates in both could mean fewer students entering CS fields overall.

The math skills issue may reflect shifts in how secondary schools teach mathematics, especially as students increasingly use calculators and digital tools for calculations. There is a mismatch here: students arrive expecting to use tools for the math, but computer science courses demand that students understand the math itself—especially for designing algorithms and analyzing how fast they run.

How AI Has Changed Assessment

AI coding assistants like ChatGPT have upended how computer science courses can test what students actually know. Take-home programming assignments, once thought to be safe from cheating, can now be completed by AI tools from a simple problem description.

The challenge goes beyond just code generation. Spotting the line between legitimate tool use and academic misconduct is now much harder. AI can now explain concepts, help with debugging, and even help students prep for exams. Distinguishing between a student using AI as a study aid and a student having AI do the work for them is not straightforward.

The faculty petition signals that this is not a Berkeley-only problem. Other universities are likely facing similar pressures. AI has arrived at a moment when most educational systems were not designed to handle it, and individual schools cannot solve this alone.

Where This Goes From Here

High failure rates like these are probably a temporary state rather than the new normal. History suggests that schools do adapt when technology disrupts how they teach and test. The real question is whether adaptation happens fast enough to keep education quality high while making sure computer science remains accessible.

Possible solutions on the table include proctored coding assessments, focusing exams on whether students understand concepts rather than just whether they can write code, and actually building AI tools into the curriculum instead of banning them. Some universities are experimenting with separate "AI-allowed" and "AI-free" assessment categories—a nod to the fact that real software development will increasingly involve AI.

The Berkeley numbers give us early real-world evidence of how generative AI is reshaping how computer science gets taught. While it looks disruptive in the short term, the historical track record suggests new ways of teaching and testing will eventually emerge as both students and faculty get comfortable with AI-assisted learning. The key will be making sure that students still develop core problem-solving skills and genuine computational thinking in the process.