1. cheat-sheets
  2. /model picker

Model Picker

Last reviewed: June 2026

Model names and pricing change frequently. Verify against provider docs before production.

Full guides: Cost, Latency, and Tokens · Choosing a Tool · Tokens and Context

Production API Tiers (2026)

TierExamplesUse when
Fast / cheapClaude Haiku, GPT-4o-mini, Gemini FlashClassification, routing, simple Q&A, high volume
BalancedClaude Sonnet, GPT-4oMost chat, coding help, tool calling
Premium / reasoningClaude Opus, o-series, GPT-4.1Hard bugs, architecture, low-volume critical tasks

Default: balanced tier. Escalate to premium only when balanced fails twice or stakes are high.

Task → Model (Production)

TaskModel tierNotes
Intent classification / routingFastOne-line label output
Customer support FAQBalancedAdd RAG for grounding
Code generation in productBalancedCap maxTokens; verify output
Complex analysis / legal / medicalPremium + human reviewNever auto-act
EmbeddingsEmbedding modelNot chat models
Batch summarization (offline)Fast or Batch APICheaper than interactive

Dev Tooling (Cursor, Copilot, Claude Code)

TaskMode / modelNotes
Tab completionsDefault / fastAccept line by line
Explain codeAsk + balancedAttach @Files
Multi-file featurePlan → Agent + balanced/premiumApprove plan first
Stuck bug (2+ failures)Debug or premiumPaste full error output
Boilerplate scaffoldBalanced or fastVerify imports after
Security-sensitive codeHuman-ledAI drafts only; strict review
Docs / commentsFast or balancedVerify facts

When to Switch Models Mid-Session

SignalSwitch to
Repeated wrong API hallucinationsPremium + @Docs or official docs pasted
Agent over-refactorsStop agent; Plan mode with file boundaries
Simple rename across filesFast model or inline Copilot
Architecture decisionAsk mode premium; you decide and implement

Cost Quick Math

Monthly cost ≈ (requests × avg_input_tokens × input_price)
             + (requests × avg_output_tokens × output_price)
TacticSaves
Truncate chat historyInput tokens
maxTokens capOutput tokens
Prompt caching (Anthropic/OpenAI)Repeated system prompt
Fast model for routingRoutes 80% of traffic cheaply
Rate limits per userAbuse / runaway loops

Provider Cheat Sheet

ProviderStrong atWatch out for
AnthropicLong context, tool use, codingModel ID strings change
OpenAIEcosystem, vision, embeddingsPricing tiers shift
Google GeminiMultimodal, GCP integrationRegion availability
Local (Ollama)Privacy, offlineQuality vs cloud frontier