Claude Models Explained: Opus, Sonnet & Haiku

Anthropic organizes Claude’s AI models into three tiers: Opus, Sonnet, and Haiku. Each tier is designed for different use cases — from complex reasoning and long coding tasks to fast everyday responses. This guide explains what each model does, who it is for, and how app and API users access them differently.

Note: Model names, capabilities, and pricing change over time. The information in this guide reflects officially published Anthropic data at the time of writing. Check Anthropic’s model overview for the current lineup.

The Claude Model Family

Claude models are grouped into three named tiers that reflect their position on the speed-vs-capability spectrum:

• Opus — the most capable tier. Designed for the hardest problems: complex reasoning, long-context analysis, and agentic coding tasks.
• Sonnet — the balanced tier. Strong intelligence with faster responses. Good for most professional and creative tasks.
• Haiku — the fastest tier. Near-frontier intelligence at the highest speed. Good for simple tasks, quick lookups, and cost-sensitive applications.

Each generation of Claude updates the models within these tiers. The current generation is Claude 5.5, with Claude Opus 5.5 (September 22, 2026) and Claude Sonnet 5.5 (September 28, 2026) as the latest tier models, and Haiku 4.5 as the fastest until Haiku 5.5 arrives. In June 2026, Anthropic added a new tier above Opus: Claude Fable 5, then its most capable widely released model (succeeded by Fable 5.1 in September 2026).

Claude Models: Current Lineup at a Glance

0

Model	Claude Opus 5.5	Claude Sonnet 5.5	Claude Haiku 4.5
Best for	Recommended default: agentic coding, knowledge work	Speed + agentic intelligence	Fast responses, simple tasks
Latency	Moderate	Fast	Fastest
Context window	1 million tokens	1 million tokens	200,000 tokens
Max output	128,000 tokens	128,000 tokens	64,000 tokens
API input price (per MTok)	$4	$2	$1
API output price (per MTok)		$10	$5
Knowledge cutoff (reliable)	June 2026	June 2026	February 2025

API pricing and model specs are verified from Anthropic’s official model overview at time of writing. Check the official page for current pricing — it can change. Claude Sonnet 5.5 keeps the $2 / $10 per MTok price Anthropic made permanent for Sonnet 5 in August 2026 (see Sonnet 5 pricing). Claude Opus 5.5 lowered Opus pricing from $5 / $25 to $4 / $20.

Claude Fable 5.1 and Claude Mythos 5.1: The Top Tier

On September 1, 2026, Anthropic released Claude Fable 5.1 (claude-fable-5-1), its top-tier model, and Claude Mythos 5.1, the same model with relaxed safeguards for invited organizations. Fable 5.1 keeps Fable 5’s $10 / $50 per MTok price but cuts cache reads 75% to $0.25 / MTok, has a June 2026 knowledge cutoff, and at launch it led Opus 5 on every benchmark Anthropic published. Three weeks later, Claude Opus 5.5 outscored it on every row of Anthropic’s own launch table at 60% lower prices — 66.4% vs 55.8% on Terminal-Bench 4.0, for example. Anthropic’s guidance now: start with Opus 5.5 for most workloads, and use Fable 5.1 for demanding reasoning and long-horizon agentic work, or when Opus 5.5 at higher effort still falls short. See Opus 5.5 vs Fable 5.1 and the Fable 5.1 launch roundup.

Claude Fable 5 and Mythos 5 (legacy)

On June 9, 2026, Anthropic launched Claude Fable 5 (claude-fable-5) — then its most capable widely released model, positioned above the Opus tier. It was superseded by Fable 5.1 on September 1, 2026 and is now a legacy model (retirement no sooner than June 9, 2027). Fable 5 is priced at $10 input / $50 output per MTok, has a 1M token context window with 128k max output, and uses adaptive thinking (always on; it does not support manual extended thinking). It is generally available via the Claude API and major cloud platforms — restored on July 1, 2026 after a brief suspension under U.S. export controls. From July 20, 2026 it was included in Max and Team Premium plans at up to 50% of weekly usage limits, while Pro and Team Standard users paid for it with usage credits at API rates (plus a one-time $100 credit from Anthropic). That in-plan slot now belongs to Fable 5.1: Max includes it, Pro does not. See the Claude Fable 5 guide for the full story, or the Opus 5 vs Fable 5 comparison to choose between the two flagships.

Claude Mythos 5 (claude-mythos-5) is the same underlying model with some safeguards lifted. It is not a public self-serve model — access is limited to approved customers in Anthropic’s Project Glasswing trusted-access program. When Fable 5’s safety classifiers trigger (in under 5% of sessions, per Anthropic), the request is routed to a fallback model instead — Claude Opus 5 since August 2026, when Anthropic also retrained the biology classifier and cut biology-related fallbacks by roughly 85%.

Claude Opus 5.5: The Recommended Default

Claude Opus 5.5, released September 22, 2026, is the model Anthropic tells developers to start with for most workloads. It costs $4 / $20 per MTok — 20% less than Opus 5 — and Anthropic estimates typical workloads cost about 40% less once its token efficiency is counted. On Anthropic’s launch benchmarks it outscored Fable 5.1 on every row, including 66.4% vs 55.8% on Terminal-Bench 4.0, and it is the default model in Claude Code on every paid plan.

Use Claude Opus 5.5 for:

• Agentic coding, large refactors, and multi-file features
• Research and knowledge work across documents, spreadsheets, and slides
• Long-context analysis across its 1M token window
• Multi-step agents that run with little supervision
• Enterprise workloads that need zero data retention

Its API default effort is medium — raise it for hard steps — and an optional fast mode runs up to 2.5× faster at $8 / $40 per MTok. Full details in the Claude Opus 5.5 guide; for the top of the range, see Opus 5.5 vs Fable 5.1.

Claude Opus 5 (legacy): released July 24, 2026 at $5 / $25 per MTok, Opus 5 is now a legacy model available until at least July 24, 2027 — see the Claude Opus 5 guide. The older Claude Opus 4.8 also remains available and handles cybersecurity requests routed from Opus 5.5.

Claude Sonnet 5.5: Best Balance of Speed and Intelligence

Claude Sonnet 5.5, released September 28, 2026, keeps Sonnet 5’s $2 / $10 per MTok price while generating output more than 30% faster, and Anthropic says it can cut the cost of a task by up to 30% by using fewer tokens and tool calls. On Anthropic’s benchmarks it nearly matches Opus 5.5 on knowledge work (1,844 vs 1,846 on GDPval-AA v2.1) and computer use (80.1% vs 81.8% on OSWorld 2.1), at half the price.

Use Claude Sonnet 5.5 for:

• Everyday writing, research, and analysis
• Coding, debugging, and code review
• Documents, spreadsheets, and summarization
• High-volume API workloads and agent pipelines
• Latency-sensitive apps, using the between_tools thinking setting

Opus 5.5 still leads on the hardest coding and reasoning tests. See Sonnet 5.5 vs Opus 5.5 for the row-by-row comparison.

Claude Sonnet 5 (legacy): released June 30, 2026, Sonnet 5 is now a legacy model available until at least June 30, 2027 — see the Claude Sonnet 5 guide.

Claude Haiku 4.5: Fastest Responses

Claude Haiku 4.5 is the fastest Claude model with near-frontier intelligence. It is designed for applications where speed and cost efficiency are more important than maximum capability.

Use Claude Haiku 4.5 for:

• Quick lookups and simple Q&A
• High-volume applications where API cost per request matters
• Real-time or near-real-time response applications
• Lightweight summarization or classification tasks
• Situations where latency is critical and the task does not require deep reasoning

Haiku’s 200k token context window is still generous for most tasks. Its API cost ($1 input / $5 output per MTok) makes it significantly more affordable than Sonnet or Opus at scale. Anthropic says Claude Haiku 5.5 will follow in the coming weeks.

App Users vs API Users: Different Access Paths

Claude is available in two ways, and each has its own pricing structure:

Claude App (Consumer Plans)

If you use Claude at claude.com, the Claude mobile app, or the Claude desktop app, you are on a consumer plan:

• Free plan: $0 — access to Claude with usage limits
• Pro plan: $17/mo (annual) or $20/mo monthly — more usage, access to multiple models
• Max plan: from $100/mo — 5x or 20x Pro usage

On consumer plans, Anthropic determines which models are available at which tier. You do not select a model by API ID — you choose it in the interface when it is offered.

Claude API (Developer Plans)

Developers who build applications using the Anthropic API pay per token (input and output) based on which model they call. API pricing uses the model IDs in the table above.

• API users select models by ID: claude-opus-5-5, claude-sonnet-5-5, claude-haiku-4-5, claude-fable-5-1
• You pay per use — no monthly subscription, though Anthropic may require adding API credits
• API pricing is separate from and not interchangeable with consumer plan pricing

See the Claude API Key guide for getting started with the API.

Which Claude Model Should You Use?

For most people the choice is between Claude Opus 5.5 and Claude Sonnet 5.5. Anthropic recommends starting with Opus 5.5, which is also Claude Code’s default model; Sonnet 5.5 comes close on many tasks at half the API price and responds faster. For quick tasks, Haiku 4.5 is the fastest option.

• Complex work and Anthropic’s default recommendation: Claude Opus 5.5
• Everyday work, high-volume, or latency-sensitive use: Claude Sonnet 5.5
• Demanding reasoning or long-horizon agents where Opus 5.5 at high effort still falls short: Claude Fable 5.1
• Quick tasks or cost-sensitive API use: Claude Haiku 4.5
• Free plan users: the Free plan includes Sonnet and Haiku models; Opus and Fable require a paid plan

If you are uncertain, start with Sonnet 5.5 for everyday tasks and switch to Opus 5.5 when you need more capability — or get a recommendation from the Claude model selector.

Comparing Claude to other AI tools? See the Claude vs GPT-4o comparison for a task-based guide, or the Claude vs ChatGPT guide for a broader product comparison.

To help choose the right model for your use case, try the Claude Model Selector tool. For API users, the Claude API Cost Estimator helps project costs before scaling. For tasks with repeated context, see the Claude Prompt Caching Guide.