Introduction: How AI-based feedback has evolved in 2026
Today, teachers have access to cutting-edge language models that provide real-time responses, adapt tone, and offer practical suggestions. Platforms like Google Gemini 3.7 Flash and Z.ai GLM-5.3 excel in advanced reasoning, making them ideal for analyzing essays, projects, or quiz answers and generating constructive feedback that sounds as if written by a human teacher.
Which LLM should you choose for school assessments?
New model versions differ in speed, cost, and specialization.
- Gemini 3.7 Flash
- GLM-5.3
- AMIE
Select a model based on class size, need for multimodal processing, and budget.
Quick comparison
| Model | Speed | Price per 1M tokens | Strengths |
|---|---|---|---|
| Gemini 3.7 Flash | Sub-second | $0.75 | Immediate multimodal feedback |
| GLM-5.3 | 2-3 seconds | $0.90 | Complex tasks and in-depth analysis |
| AMIE | 3-4 seconds | $1.20 | Interactive conversation and tutoring |
How to create an effective feedback prompt
A good prompt tells the model what to evaluate, who the audience is, and what style to use.
Example prompt for a writing assignment:
Generate detailed feedback for this essay:
"[ESSAY TEXT]"
Instructions:
- Highlight strengths and areas for improvement.
- Provide concrete suggestions and review grammar.
- Adapt the tone to the school level (e.g., high school vs. university).
- Use encouraging and specific language.This prompt works because it specifies the output, audience, and tone, ensuring consistent and useful responses.
Integrating LLMs into an existing LMS
Most modern learning management systems (LMS) support APIs that can be connected to LLM services. Here’s a quick guide:
- Register an API key with your chosen provider (Google Cloud, Z.ai, etc.).
- Use the LMS plugin or connector to automatically send student work to the model.
- Retrieve the feedback and publish it in the grade book or a dedicated discussion.
Plugins likeedX-AI-FeedbackorMoodle LLM Assistantsimplify this workflow.
Example Moodle plugin integration
// Hook to send essay text to LLM upon submission
def send_to_llm(instance_id, student_work):
url = "https://api.google.ai/v1/models/gemini-3.7-flash:generateContent"
headers = {"Authorization": "Bearer " + GOOGLE_API_KEY}
payload = {
"contents": [{
"parts": [{
"text": f"Generate feedback for this essay:\n\n{student_work}\n\nInstructions:\n- Highlight strengths and areas for improvement.\n- Provide concrete suggestions.\n- Adapt the tone to the school level."
}]
}]
}
response = requests.post(url, headers=headers, json=payload)
return response.json()['candidates'][0]['content']['parts'][0]['text']Real-world use cases in 2026
High schools in Milan.They adopted Gemini 3.7 Flash to evaluate Italian essays, reducing correction time by 70% and allowing teachers to focus more on personalized tutoring.
They use GLM-5.3 to evaluate large programming projects, receiving detailed assessments on efficiency, readability, and adherence to standards.
AMIE has been integrated into a communication lab to generate realistic feedback during medical visit simulations, helping students improve empathy and clinical language.
Best practices for ethical and effective feedback
- **Transparency:** Always inform students when an LLM has contributed to the evaluation.
- **Human review:** Use AI as a draft, then apply expert teacher judgment.
- **Data security:** Encrypt student text before sending and use only GDPR/CCPA-compliant APIs.
- **Cultural adaptation:** Customize feedback language to the cultural context of the class.
Key takeaways
- Choose an LLM based on speed, cost, and task complexity.
- Write clear prompts that specify audience, tone, and required actions.
- Integrate models via LMS APIs to automate the grading process.
- Always combine AI feedback with human review to ensure quality and fairness.
- Stay updated on new releases (like Gemini 3.7 Flash and GLM-5.3) to leverage the latest features.