← d3dev

Anthropic Claude Models — Full Landscape (March 2026)

At a Glance

Anthropic currently offers 12 Claude models across 4 generations (Claude 3 through Claude 4.6). The latest flagship is Claude Opus 4.6 — their most intelligent model, purpose-built for agents and coding. Two models (Sonnet 3.7, Opus 3) are officially deprecated but still listed for pricing. The active lineup spans from $0.25/MTok (Haiku 3) to $5/MTok input (Opus 4.6).

Anthropic Claude Models — Full Landscape (March 2026)

FieldValue
TitleAnthropic Claude Models — Active Model Roster & Pricing
SourceAnthropic Official Documentation
Linkhttps://docs.anthropic.com/en/docs/about-claude/models
Date PublishedOngoing (docs updated continuously)
Date Downloaded2026-03-05
Tagsanthropic, claude, models, pricing, AI, LLM

At a Glance

Anthropic currently offers 12 Claude models across 4 generations (Claude 3 through Claude 4.6). The latest flagship is Claude Opus 4.6 — their most intelligent model, purpose-built for agents and coding. Two models (Sonnet 3.7, Opus 3) are officially deprecated but still listed for pricing. The active lineup spans from $0.25/MTok (Haiku 3) to $5/MTok input (Opus 4.6).

Sam's TLDR

Anthropic's model lineup is layered by generation and by speed/intelligence tradeoff within each generation. The current recommendation is Opus 4.6 for complex tasks and Haiku 4.5 for speed-sensitive work, with Sonnet 4.6 as the balanced middle. Older models (Opus 4, 4.1, Sonnet 4, etc.) are still available and priced identically or higher than their successors — the 4.6 generation is actually cheaper per-token than the 4.0/4.1 Opus models ($5 vs $15/MTok input). New features like extended thinking, adaptive thinking, fast mode, and 1M context windows are exclusive to the latest generation.

Active Models — Quick Reference

ModelAPI IDInput $/MTokOutput $/MTokContextMax OutputKnowledge CutoffBest For
Claude Opus 4.6claude-opus-4-6$5$25200K / 1M (beta)128KMay 2025 (reliable), Aug 2025 (training)Agents, coding, complex reasoning, multi-hour research
Claude Sonnet 4.6claude-sonnet-4-6$3$15200K / 1M (beta)64KAug 2025 (reliable), Jan 2026 (training)Code gen, data analysis, agentic tool use, balanced speed+intelligence
Claude Opus 4.5claude-opus-4-5$5$25200KPrevious flagship, same pricing as 4.6
Claude Opus 4.1claude-opus-4-1$15$75200KLegacy premium tier
Claude Opus 4claude-opus-4$15$75200KLegacy premium tier
Claude Sonnet 4.5claude-sonnet-4-5$3$15200K / 1M (beta)64KPrevious balanced model
Claude Sonnet 4claude-sonnet-4$3$15200K / 1M (beta)64KEarlier balanced model
Claude Haiku 4.5claude-haiku-4-5$1$5200K64KFeb 2025 (reliable), Jul 2025 (training)Real-time apps, high-volume processing, sub-agents, cost-sensitive
Claude Haiku 3.5claude-haiku-3-5$0.80$4200KBudget option, still capable
Claude Haiku 3claude-haiku-3$0.25$1.25200KCheapest option, simple tasks

Deprecated (Still Priced, Being Phased Out)

ModelInput $/MTokOutput $/MTok
Claude Sonnet 3.7$3$15
Claude Opus 3$15$75

Key Points

Model Selection Guide

NeedModelWhy
Maximum intelligenceOpus 4.6Best reasoning, coding, agents
Speed + intelligence balanceSonnet 4.6Fast, capable, great for most tasks
Lowest latencyHaiku 4.5Fastest model, near-frontier intelligence
Lowest costHaiku 3$0.25/MTok, good for simple tasks
Most recent knowledgeSonnet 4.6Training data through Jan 2026
Batch processingAny model + Batch API50% discount on all tokens

Full Summary

Anthropic's model lineup as of March 2026 reflects a mature, multi-generational portfolio. The Claude 4.6 generation (Opus and Sonnet) represents the current state of the art, with significant improvements in intelligence, coding capability, and agentic performance over prior generations.

The pricing structure reveals an interesting pattern: newer doesn't always mean more expensive. Claude Opus 4.6 costs $5/MTok input — the same as Opus 4.5 and a third of what Opus 4.0 and 4.1 cost ($15/MTok). This suggests Anthropic has achieved substantial efficiency gains in training and inference, passing those savings on while also keeping legacy models at their original price points.

The three-tier structure (Opus/Sonnet/Haiku) has remained consistent since Claude 3. Each tier targets a different optimization point: Opus for raw capability, Sonnet for the speed-intelligence sweet spot, and Haiku for throughput and cost. Within each tier, newer generations strictly dominate older ones — there's no reason to use Opus 4 over Opus 4.6 unless you have specific compatibility requirements.

Notable new capabilities exclusive to the 4.6 generation include adaptive thinking (the model dynamically adjusts reasoning depth), fast mode for Opus (2.5x speed at premium pricing), and 128K max output tokens for Opus. The 1M context window, initially introduced with earlier models, is now available on both 4.6 models.

The deprecated models (Sonnet 3.7 and Opus 3) are still accessible and priced but marked for phase-out. Developers should migrate to their respective successors.

For Sam specifically: we run on Claude Opus 4.6 (Thinking) — the top of the lineup. The 128K max output and extended thinking make it ideal for the kind of deep, multi-step work we do.

References

  1. [1]Anthropic — Models Overview. https://docs.anthropic.com/en/docs/about-claude/models
  2. [2]Anthropic — Choosing the Right Model. https://docs.anthropic.com/en/docs/about-claude/models/choosing-a-model
  3. [3]Anthropic — Pricing. https://docs.anthropic.com/en/docs/about-claude/pricing