claude-46

name: claude-4.6 description: | Claude Opus 4.6 and Sonnet 4.6 — Anthropic's frontier models with adaptive thinking, 1M context, agent teams, fast mode, and computer use. Use when: claude, opus, sonnet, adaptive thinking, agent teams, fast mode, computer use, claude code. role: Specialist Executor intent: Route tasks to the optimal Claude model variant with correct thinking effort and feature flags kpis: [model_selection_accuracy, cost_per_task, adaptive_effort_hit_rate] status: active priority: critical triggers:

• claude
• claude opus
• claude sonnet
• opus 4.6
• sonnet 4.6
• adaptive thinking
• agent teams
• fast mode
• computer use
• claude code execution: target: internal route: "" dependencies: env:
  • OPENROUTER_API_KEY apis:
  • https://openrouter.ai/api/v1
  • https://api.anthropic.com/v1 released: "2026-02-05"

Claude 4.6 Skill

Claude Opus 4.6: Released February 5, 2026 Claude Sonnet 4.6: Released February 17, 2026

Model Lineup

Core Workflow — Model Routing

1. Classify incoming task by complexity and latency requirement
2. Select model + thinking effort per the Routing Rules below
3. Apply feature flags (fast mode, 1M context, computer use) as needed
4. Execute via Anthropic API or OpenRouter
5. Log model selection, effort level, and token usage

Adaptive Thinking (Replaces Extended Thinking)

No more manual budget_tokens. The model dynamically decides reasoning depth.

const response = await anthropic.messages.create({
  model: "claude-opus-4-6",
  max_tokens: 16000,
  thinking: { type: "adaptive" },
  messages: [{ role: "user", content: "..." }],
});

Routing Rules for AIMS

IF task is core orchestration (deploy, provision, scale)
  → claude-opus-4-6, adaptive thinking, effort: high or max

IF task is code review, PR analysis, debugging
  → claude-sonnet-4-6, adaptive thinking, effort: medium

IF task is classification, intent detection, routing
  → claude-haiku-4.5 or gemini-3.0-flash (cheaper, faster)

IF task needs real-time voice response
  → claude-opus-4-6 fast mode (2.5x speed)

IF task is bulk content generation (overnight)
  → claude-sonnet-4-6 batch API (50% off)

IF task is security audit
  → claude-opus-4-6, adaptive thinking, effort: max

IF task is UI testing / computer use
  → claude-sonnet-4-6 with computer_use tool

Key Features

See references/api-patterns.md for detailed API usage including:

• 1M context window (beta header)
• 128K max output (Opus)
• Fast mode (2.5x speed, 6x pricing)
• Compaction API (infinite conversations)
• Agent Teams (parallel sub-agents)
• Computer Use (72.7% OSWorld)
• Prompt caching (90% savings on cache reads)
• Batch API (50% discount)

Anti-Patterns

• Do NOT use budget_tokens on 4.6 models — use adaptive thinking instead
• Do NOT prefill assistant messages on Opus 4.6 — returns 400 error
• Do NOT use Fast Mode for batch workloads — 6x pricing defeats the purpose
• Do NOT default to Opus for everything — Sonnet 4.6 beats previous Opus 4.5 in 59% of cases
• Do NOT ignore prompt caching — cache reads are 90% cheaper than fresh input
• Do NOT skip the max effort level for critical decisions — it's exclusive to Opus 4.6

Quality Gates

• Model selection matches task complexity (don't use Opus for simple routing)
• Prompt caching enabled for repeated system prompts
• Batch API used for non-urgent bulk workloads
• Fast mode reserved for latency-critical voice/realtime scenarios
• Token usage logged for cost tracking

Hooks

• trigger: Claude model invocation required
• pre_gsd: Validate API key, select optimal model + effort, check rate limits
• post_gsd: Log model, effort, tokens consumed, latency

Limits

• Opus 4.6: 200K context (1M with beta header)
• Sonnet 4.6: 200K context (1M with beta header)
• Rate limits vary by tier (see Anthropic docs)
• Fast mode: research preview, may have availability constraints