RAG over your docs
Search by meaning, keep the best passages, answer only from them.
- 01Embedvoyage-4
- 02RerankRerank 4 Pro
- 03AnswerClaude Haiku 5.5
A field guide to AI models
Every model gets a lesson: what it's good at, three real tasks, where it sits in your stack, and the request to paste.
NEGGING FIELD GUIDE · READY592 models · 20 lessons · 7 boards
how can I use Claude Opus 5.5 in my Next.js project?
Claude Opus 5.5 · AnthropicWHATAnthropic's flagship for demanding reasoning, multi-step changes in large codebases, and long-horizon agent work.STACKPlan & reason · the planner in a coding agentPut Opus in the main agent loop.GOOD AT+ Multi-step changes across a large codebase+ Long-horizon agent loops that call tools for many turns+ Hard debugging and design reviews where you want the careful answerCALLPOST /api/v1/chat/completionsmodel "anthropic/claude-opus-5.5" + messages, reasoning, toolsLIVE1.0M context · $4 in · $20 out /Mopen the full lesson →The board
Pick a job. See which models do it, then open a lesson to learn how to use one.
Which model should run your agent loop?
Frontier models that plan, call tools, and edit code across many steps.
Intelligence index· Artificial Analysis via OpenRouter
Claude Opus 5.5
GPT-6.1 Sol
Claude Sonnet 5.5
Gemini 4 Argon
Grok 4.7
GLM 5.3
Kimi K3
| # | Model | Best at | AA index | $/M in · out | Context | |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 Anthropic | The planner in a coding agent. Multi-step changes across a large codebase | 59.1 | $4 · $20 | 1.0M | Lesson |
| 2 | GPT-6.1 Sol OpenAI | The document-heavy agent. Agentic coding with tool calls | 54.6 | $2 · $10 | 1.1M | Lesson |
| 3 | Claude Sonnet 5.5 Anthropic | The everyday builder. Building a feature from a clear ticket | 54.5 | $2 · $10 | 1.0M | Lesson |
| 4 | Gemini 4 Argon | The brain when inputs are not just text. Reasoning over mixed media: video, audio, screenshots, and PDFs together | 50.4 | $2 · $10 | 1.0M | Lesson |
| 5 | Grok 4.7 xAI | The self-checking engineer. Long-running software engineering tasks | 35.5 | $2 · $6 | 500K | Lesson |
| 6 | GLM 5.3 Z.ai· open weights | The open-weight agent. Long-horizon coding agents on an open-weight model | 31.7 | $0.070 · $1.4 | 1.0M | Lesson |
| 7 | Kimi K3 Moonshot· open weights | The open-weight multimodal agent. Open-weight agents that also need to see images or video | 17.4 | $0.192 · $16 | 1.0M | Lesson |
AA index = Artificial Analysis Intelligence Index, as published on each model's OpenRouter listing. A dash means the listing has no score. Prices are per million tokens, live from OpenRouter.
How a stack fits together
Five layers. Most apps use three.
A model is one part of a system. Learn the layers once, and every new release has an obvious place to go.
Layer 01
Retrieve
Find the right context: embeddings and rerankers.
Layer 02
Plan & reason
The main model: plans, calls tools, writes the hard parts.
Layer 03
Do the volume
Fast models for subagents, extraction, and bulk text.
Layer 04
Gate & route
Decision models return probabilities your code thresholds.
Layer 05
Draw, speak, listen
Image, speech, and transcription models at the edges.
Recipes
Whole stacks you can copy.
Each one follows an OpenRouter cookbook, so the request shapes match the docs.
Coding agent with a safety gate
A planner edits code, subagents do the legwork, a decision model checks risky tool calls.
- 01PlanClaude Opus 5.5
- 02SubagentsClaude Haiku 5.5
- 03GateJev 1.14
Push-to-talk voice assistant
Hold a button, ask out loud, hear the answer.
- 01ListenScribe v2
- 02ThinkClaude Haiku 5.5
- 03SpeakEleven v4 Turbo
Tell us the job. Get a stack.
Describe what you're building. We'll suggest models for each layer and link the lessons that show you how.
Find my model