Add prompting guide for next-gen models: Opus 4.6, Sonnet 4.6, Qwen 2.5, GLM-5

[Copilot]

Adds practical prompting strategies for latest-generation LLMs tailored to data science workflows.

Changes

New notebook: notebooks/03_next_gen_model_prompting.ipynb

• Prompt templates for 7 DS task categories: dataset analysis, metrics design, fraud detection, HITL systems, dataset curation, ML infrastructure, LLM evaluation
• Model-specific guidance with selection decision tree
• Advanced techniques: chain-of-thought, few-shot examples, LLM-as-judge patterns

Documentation updates: docs/generative-ai.md

• Added Qwen 2.5 and GLM-5 to model catalog
• New "Interactive Tutorials" section linking all notebooks
• Fixed: "Clause" → "Claude"

Example

Fraud analysis prompt with structured output requirements:

prompt = """
Analyze chargeback spike: 40% increase in $50-$100 range, 2-4 AM EST.
Affected: gaming/digital goods, 12 of 450 merchants, last 14 days.

Tasks:
1. Hypothesize 3 root causes (ranked by likelihood)
2. Design investigation plan with specific queries
3. Propose immediate mitigation + long-term prevention
4. Recommend additional data points to collect
"""

Model selection logic: Opus 4.6 for novel patterns, Sonnet 4.6 for production batch processing, Qwen 2.5 for on-premise/privacy-sensitive, GLM-5 for bilingual teams.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Add prompting examples for next-gen models: Opus 4.6, Sonnet 4.6, Qwen 2.5, GLM-5</issue_title>
<issue_description>It would be beneficial to have dedicated notebooks or documentation showcasing effective prompting strategies for the latest model releases. Specifically Opus 4.6, Sonnet 4.6, the Qwen 2.5 series, and GLM-5. This will help users leverage the specific capabilities of these newer architectures.</issue_description>

<agent_instructions>Effective prompting strategies as they relate to typical data science tasks such as: analyze complex datasets, design metrics, build dashboards, Develop and track metrics to measure and monitor the performance of our risk products and the effectiveness of risk management strategies
Conduct in-depth analyses to uncover valuable insights contributing to fraud reduction and higher approval rates for our clients, Designing human-in-the-loop systems
Establishing dataset collection and curation methodologies, Architecting AI infrastructure (model serving, monitoring, scaling)
Optimizing performance and defining quality metrics, eployed evaluation frameworks for LLMs, with an understanding on performance, reliability, and bias assessment</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• Fixes Add prompting examples for next-gen models: Opus 4.6, Sonnet 4.6, Qwen 2.5, GLM-5 #10

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.