Claude: Anthropic's AI Assistant and the Complete 5.5 Model Family (2026)
Quick verdict
Claude is Anthropic's assistant, now built on the Claude 5.5 family: Haiku 5.5 at $0.10/$0.50 per million tokens for high-volume work, Sonnet 5.5 at $2/$10 as the everyday workhorse, Opus 5.5 at $4/$20 as the recommended default, and Fable 5.1 at $10/$50 for the hardest reasoning. All current models share a 1M-token context window and 128K max output.
What Claude Is Now
Anthropic's assistant is no longer a single model with a couple of variants. As of October 2026, Claude is a five-tier family, and the tiers differ more in price than in capability ceiling:
| Model | Input / Output (per 1M) | Cache reads | Best for |
|---|---|---|---|
| Haiku 5.5 | $0.10 / $0.50 (≤100K) | $0.01 | Classification, extraction, routing, compaction, subagents |
| Sonnet 5.5 | $2 / $10 | $0.10 | Everyday coding, documents, slides, spreadsheets |
| Opus 5.5 | $4 / $20 | $0.20 | Anthropic's recommended default for most work |
| Fable 5.1 | $10 / $50 | $0.25 | Demanding reasoning, long-horizon agentic work |
| Mythos 5.1 | Trusted access | — | Fable 5.1 with more permissive safeguards |
Every current model shares a 1M-token context window with 128K maximum output, and a knowledge cutoff of June 2026. The full spread from cheapest to most expensive tier is 100x — for the same context window.
What Each Tier Actually Does
Haiku 5.5 is the tier that changed the family's economics in October 2026. It costs a tenth of Haiku 4.5 on the same prompt shape, carries five times the context window, and is the first Haiku-class model with an adjustable effort dial (Low through Max, defaulting to Medium). Cache reads are $0.01 per million — the deepest cache discount Anthropic has published.
Sonnet 5.5 is the workhorse: fastest and most balanced for well-scoped work, with cache reads halved to $0.10 in October, which Anthropic estimates makes agentic tasks roughly 20% cheaper. It tops the Coding Agent Index at maximum effort — at a real cost of up to $14.19 per task, so effort settings decide the bill.
Opus 5.5 is what Anthropic suggests starting with. It leads the Artificial Analysis Intelligence Index at 58, scores 66.4% on Terminal-Bench 4.0, and beats GPT-6 Astra on FrontierCode at roughly a fifth of the cost per task at default effort. It is the default model in Claude Code on Pro, Max, Team and Enterprise plans.
Fable 5.1 remains the ceiling for reasoning-heavy and long-horizon work, with cache reads at a quarter of its input rate — which makes it cheaper than its predecessor for any workload that reuses a long prefix.
Key Capabilities
Adaptive thinking with effort control. Every 5.5-family model reasons before answering, and effort settings — not manual token budgets — govern how much. Opus-class models cannot switch thinking off; the effort dial is the control.
1M-token context. The entire family handles large codebases, document sets, and long agent sessions without truncation. Haiku 5.5's minimum cacheable prompt is 512 tokens, and Sonnet 5.5 dropped from 1,024 to 512.
Prompt caching that materially changes bills. In agent loops, cached context — system prompts, tool definitions, repository instructions — dominates token consumption. The October cache cuts matter more to total spend than headline rates do.
Computer use and browser use. The Claude SDKs added beta support for both, and Haiku 5.5's speed-price combination makes it a natural fit for those tasks.
Alignment and safeguards. The 5.5 family reports the strongest behavioral audit results Anthropic has measured, with tiered safeguards for cyber and biology work and verification programs for organizations that need broader access.
Pricing Beyond the API
Claude Code defaults to Opus 5.5 on Pro, Max, Team and Enterprise plans, and Anthropic raised five-hour usage limits by 20% at the Opus 5.5 launch. Max and Team subscribers now also receive monthly Platform API credits — $100 for Max 5x, $200 for Max 20x, and up to $500 pooled across Team users — usable on any model including Haiku 5.5.
One deadline worth noting: Claude Sonnet 4.5 was deprecated on September 30, 2026 and retires on November 30, 2026. Teams still running it should migrate — and read the migration notes, because adaptive thinking being on by default breaks parsers that assume responses start with text.
Who Claude Is Best For
- Software engineers — the strongest coding lineup available, with a real subagent tier
- Teams running agent pipelines — five price points on one context window is an architecture, not a menu
- Writers and editors — Claude remains the most nuanced prose model in this generation
- Analysts and researchers — adaptive thinking plus 1M context handles long, careful work
The Bottom Line
Claude's 2026 lineup is best understood as one context window sold at five prices, with an effort dial at every tier. That structure is what makes multi-agent architectures practical: the cheap tier is no longer a truncation compromise.
Start with Opus 5.5 unless you have a reason not to, move high-volume work down to Haiku 5.5, and let Sonnet 5.5 carry the middle. The pricing and capability details for each are covered in the deep dives below.
Related Articles
Frequently Asked Questions
What models does Claude include in 2026?+
How much does Claude cost?+
What is Claude's context window?+
Is Claude good for coding?+
What happened to the older Claude models?+
Pros
- Five tiers sharing one 1M-token context window across a 100x price range
- Opus 5.5 leads the Artificial Analysis Intelligence Index at 58 and is the default in Claude Code
- Haiku 5.5 is the first Haiku-class model with an adjustable effort dial
- Sonnet 5.5 cache reads halved to $0.10, cutting agentic costs by roughly 20%
- Adaptive thinking, computer use and browser use are available across the platform
Cons
- The top tier (Fable 5.1) costs 100x the cheapest tier per token
- Fable-level capability is expensive: $10/$50 per million tokens
- Mythos 5.1 remains restricted to vetted organizations through trusted access programs
- Reasoning spend is governed by effort settings that must be tuned per workload
- Haiku 5.5 pricing quintuples on prompts above 100,000 tokens
Keep reading