Claude

Anthropic’s frontier LLM family. Widely considered the best model for work tasks and coding. Distinct from Claude Code, which is the CLI agent harness built on top of Claude models. See anthropic for the company-level entity.

Claude 4 (May 2025 generation — launch pricing)

TierInputOutputNotable
Opus 4$15/1M$75/1M#1 SWE-bench Verified at launch; one-shot full-stack capability
Sonnet 4$3/1M$15/1MFaster than Opus; ~$2.60 to generate a SaaS landing page via Cline (source)

The model family later evolved into the 4.5 / 4.6 line that powers Claude Code today. Current-generation API pricing (significantly lower for Opus) is in the Advisor Strategy section below.

Strengths

  • Work tasks: Modifying Excel documents, writing Word documents, analyzing large amounts of data
  • Coding: Strong coding performance across the model family
  • Integrations and tools: Supports pluggable integrations (Gmail, Notion, Figma, Slack, HubSpot, and many more) directly from the Claude web interface
  • Skills: Users can define reusable skills that customize Claude’s behavior (e.g., a “humanizer” skill to reduce AI-sounding writing)

Tiers (as of early 2026)

PlanPriceNotable
Free$0Web, iOS, Android, desktop; code gen and writing; not the frontier model
Pro$17–20/monthClaude Code, Claude co-work, unlimited projects, research, frontier models, Excel/PowerPoint integration
Max$100/monthEverything in Pro, much more usage
Heavy$200/monthMaximum usage tier

The jump from free to Pro ($20/month) is described as significant — unlocks Claude Code and many use cases unavailable on free tier.

Advisor Strategy (Messages API, beta)

The Advisor Strategy pairs an expensive model (Opus) as an advisor with a cheaper model (Sonnet or Haiku) as the executor. The executor handles every request and only escalates to the advisor when it self-assesses a task as difficult. API-level feature — add type: "advisor_2026_03_01", name: "advisor", and optionally max_uses to the Messages API request.

API token pricing (4.5/4.6 generation):

TierInput / 1M tokensOutput / 1M tokens
Opus$5$25
Sonnet$3$15
Haiku$1$5

Benchmarks from Anthropic’s evaluations:

  • Sonnet + Opus advisor = +2.7pp on SWE-bench over Sonnet alone, −12% cost per agentic task
  • Haiku + Opus advisor = 41.2% on BrowseComp vs 19.7% solo (>2× performance)

The max_uses parameter caps advisor calls per request for budget control. Escalation is non-deterministic — the executor’s self-assessment of difficulty varies by model and prompt. Test extensively before production.

See pattern-6-advisor-strategy for the full pattern breakdown. For the Claude Code equivalent (/model opus-plan), see advisor-strategy-in-claude-code.

Source: summary-nate-herk-advisor-strategy.

Model Family

The Claude family includes multiple tiers (Haiku, Sonnet, Opus) at different capability and cost levels. See the model IDs used in Claude Code for specifics.

Claude vs Claude Code

  • Claude: The model family; conversational interface at claude.ai
  • Claude Code: A CLI agent harness that wraps Claude models with tools, file access, and agentic capabilities for software development

Compared to Other Frontier Models

According to Matthew Berman:

  • Claude: best for work and coding
  • ChatGPT: best for ease of use
  • Gemini: best for search and deep research

See Also