Working with Frontier Coding Models: Claude, GPT, and Gemini for Developers

Class Duration

7 hours of live training delivered over 1-2 days to accommodate your scheduling needs.

Student Prerequisites

Working experience with at least one programming language
Basic familiarity with REST APIs is helpful but not required

Target Audience

Software engineers and tech leads who regularly use or plan to use frontier LLMs directly in their workflows or applications. Ideal for developers who want a principled basis for choosing between Claude, GPT, and Gemini rather than defaulting to whichever model they tried first, as well as teams standardizing on a model strategy.

Description

This course takes a practical, model-agnostic approach to the frontier coding models that matter most for software development in 2026. We compare Anthropic Claude (Opus 4.7, Sonnet 4.6, Haiku 4.5 and the 1M-context tiers), OpenAI GPT (GPT-5 and GPT-5.5 family, including the o-series reasoning models), and Google Gemini (3.x family) across capability dimensions relevant to developers: code generation quality, instruction following, context length, reasoning, multimodal support, function/tool calling, structured output reliability, pricing tiers, and latency characteristics. Labs benchmark the same representative engineering tasks across all three families so participants develop their own informed intuitions. We also cover how to access these models via API, SDK, and IDE integrations, and when to reach for a smaller or specialized model instead.

Learning Outcomes

Compare Claude, GPT, and Gemini across capability dimensions relevant to professional software development.
Evaluate cost, latency, and context-length tradeoffs for different use case shapes.
Access frontier models via REST API, official SDKs, and common IDE integrations.
Select the appropriate model tier (nano/mini, standard, extended-context) for a given task.
Explain when to use specialized or fine-tuned models vs. frontier generalists.
Apply consistent evaluation criteria to new model releases.

Training Materials

Comprehensive courseware is distributed online at the start of class. All students receive a downloadable MP4 recording of the training.

Software Requirements

Students need free or trial API access to at least two of the three covered model families. Labs use lightweight scripts runnable in Python or TypeScript.

Training Topics

The Frontier Model Landscape

How frontier models are trained and evaluated
Benchmark literacy: what MMLU, HumanEval, SWE-bench, and code-agent benchmarks measure
Model families and versioning: Claude 4.x (Opus 4.7, Sonnet 4.6, Haiku 4.5), GPT-5 / GPT-5.5 / o-series, Gemini 3.x
Extended-context tiers (1M+ context) and when they matter
The rise of reasoning/thinking modes and "extended thinking"

Claude for Developers

Strengths: extended context (1M tier), instruction following, code analysis, agentic tool use
Claude API: authentication, Messages API, streaming, prompt caching
Tool use, computer use, and the Claude Agent SDK
Cost and rate limit profile across Opus, Sonnet, and Haiku tiers

GPT and OpenAI for Developers

Strengths: ecosystem maturity, function calling, fine-tuning, the o-series reasoning models
OpenAI API: chat completions, Responses API, Assistants, streaming
JSON mode, native structured outputs
The Codex model family: GPT-5.5 (flagship), the GPT-5.4 family, and the GPT-5.2-Codex specialized variant; reasoning-level selection
Codex as a multi-surface platform (CLI, IDE extensions, cloud agents) — covered in depth in the dedicated Codex in Practice course
Cost and rate limit profile

Gemini for Developers

Strengths: multimodal input, very long context, Workspace integration
Gemini API via Google AI Studio and Vertex AI
Code Execution tool and grounding with Search
Cost and rate limit profile

Model Selection Framework

Task shape vs. model capability matrix
Context-length budget planning
Latency-sensitive vs. batch-workload decisions
When to use smaller (local or fast) models
Mixing models in a pipeline

API, SDK, and IDE Integration

Direct REST API vs. official SDKs
IDE integrations: VS Code, Visual Studio, JetBrains IDEs, Zed, Cursor, and Windsurf
Managing API keys and credentials safely
Observability: logging requests and costs

Workshop

Side-by-side benchmarking lab
Model selection scenario exercises
Q&A session