Agentic WeeklyGet the weekly

Which Claude model should you use? Haiku 5.5, Sonnet 5.5, Opus 5.5 or Fable 5.1

Anthropic's own advice is to start with Claude Opus 5.5 ($4 input, $20 output per million tokens) for most work, move to Sonnet 5.5 ($2/$10) for cheaper everyday coding, use Haiku 5.5 (from $0.10/$0.50) for subagents, and keep Fable 5.1 ($10/$50) for demanding reasoning. Verified 2026-10-26; we have not run these models.

By Agentic Weekly, operated independently from Texas, US · Last updated (America/Chicago) · How we work

Checked by Agentic Weekly's editorial process: vendor docs re-read.

Last verified: 2026-10-26 (America/Chicago) · Versions covered: Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 as listed on Anthropic's models overview, read 2026-10-26 · Evidence: Reported (see “What we tested”)

Quick chooser by task

The "start with" column follows Anthropic's choosing-a-model guide and Haiku 5.5 announcement. The "move up" and "move down" columns are our editorial reading of the same pages, not test results.

Which Claude model to start with, by task (Anthropic guidance plus our editorial judgement, read 2026-10-26)
TaskStart withMove up ifMove down if
Most workloads, if you are unsureClaude Opus 5.5: Anthropic says to start here for most workloadsYour evals at higher effort still fall short: try Fable 5.1Quality is already well above what you need: try Sonnet 5.5
Everyday coding, agents and enterprise workloads at lower costClaude Sonnet 5.5: Anthropic calls it the best combination of speed and intelligenceComplex multi-hour agentic coding or large refactors: Opus 5.5The task is narrow and high-volume: Haiku 5.5
Long agent runs and large refactorsOpus 5.5: Anthropic lists multi-hour autonomous coding agents and large-scale refactoringSessions that run for hours or deep multistep research: Fable 5.1Work is well scoped and short: Sonnet 5.5
The hardest reasoning and long-horizon workClaude Fable 5.1: Anthropic's most capable model open to all customersNothing above it is open to everyoneOpus 5.5 at higher effort first; Anthropic says to move to Fable if that falls short
Writing and knowledge workOpus 5.5 or Sonnet 5.5; Anthropic lists content creation under Sonnet 5.5 and knowledge work under Opus 5.5Quality matters more than cost: Opus 5.5 or Fable 5.1Drafts and summaries at volume: Haiku 5.5
Extraction, classification, routingClaude Haiku 5.5: built for high-volume, latency-sensitive workThe task needs judgement across many files: Sonnet 5.5Already the cheapest Claude model
Subagents and compactionHaiku 5.5: Anthropic says it pairs well with Opus 5.5 and Sonnet 5.5 as a subagent on coding workThe subagent must do complex coding itself: Sonnet 5.5Not applicable

Price and context at a glance

Claude models: price, context and latency from Anthropic's models overview, read 2026-10-26 (US dollars per million tokens)
ModelInputOutputContextLatency (Anthropic)Default effortAPI id
Claude Fable 5.1$10$501M tokensSlowerhighclaude-fable-5-1
Claude Opus 5.5$4$201M tokensModeratemediumclaude-opus-5-5
Claude Sonnet 5.5$2$101M tokensFasthighclaude-sonnet-5-5
Claude Haiku 5.5$0.10 up to 100K-token prompts, $0.50 above$0.50 up to 100K-token prompts, $2.50 above1M tokensFastestmediumclaude-haiku-5-5

All four list 128K maximum output and a June 2026 knowledge cutoff on that page. Cache-read prices differ: Opus 5.5 $0.20 and Sonnet 5.5 $0.10 per million tokens (Anthropic's Opus 5.5 and Haiku 5.5 posts). Haiku 5.5 is the only one priced by prompt length; for the others Anthropic's pricing page says the full 1M context is included at standard pricing. See Anthropic pricing for batch discounts and caching.

When Sonnet 5.5 is enough

Sonnet 5.5 costs half of Opus 5.5 per token and Anthropic's models overview calls it "the best combination of speed and intelligence" and lists code generation, data analysis, content creation, visual understanding and agentic tool use as its fit. Anthropic's Haiku post says it cut Sonnet 5.5's cache-read price by 50% to $0.10 per million tokens, which it says lowers the cost of most agentic work on Sonnet 5.5 by around 20%. If your task passes your own check on Sonnet 5.5, that is a sensible reason to stay on the cheaper model.

When to pay for Opus 5.5

Anthropic's models overview says to start with Opus 5.5 for most workloads and describes it as built for long-running agentic coding and knowledge work. Its launch post claims it costs 40% less to run than Opus 5 on typical workloads; that is a vendor claim about its own previous model, not a comparison with other vendors. Anthropic also says that Opus 5.5 and Haiku 5.5 default to medium effort while Sonnet 5.5 and Fable 5.1 default to high, and that "Tuning effort is often a better lever than switching models."

When Haiku 5.5 is the right worker

Anthropic positions Haiku 5.5 for quick, repetitive, high-volume work such as summaries, compaction, database queries and classification, and for speed-sensitive jobs like live support and browser use. Its own post is careful about limits: "Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding tasks", while Haiku 5.5 is best suited to narrowly scoped tasks like compaction, summarization or subagent work. Customer quotes in that post are Anthropic-selected and we have not verified them.

Where does Fable 5.1 fit?

Only from Anthropic's pages: Fable 5.1 is the highest-priced of these four at $10/$50, described as for demanding reasoning and long-horizon agentic work. Anthropic says to move to it when your evals on Opus 5.5 at xhigh or max effort still fall short, and that it extends Fable 5 with stronger long-running agentic coding, knowledge work and research at the same prices. We list a release-date discrepancy between sources on the Fable 5.1 page rather than guess here.

Using these in Claude Code

Claude Code's docs say that on the Anthropic API the opus, sonnet and haiku aliases resolve to Opus 5.5, Sonnet 5.5 and Haiku 5.5, and that the fable alias resolves to Fable 5.1. On other providers the aliases can point to older models (for example the sonnet alias resolves to Sonnet 4.5 on Amazon Bedrock and Google Cloud's Agent Platform), so check your provider's row in that table. They also describe an opusplan mode that uses Opus during plan mode and then switches to Sonnet for execution. Start from Claude Code, and see GPT-6.1 Terra vs Claude Opus 5.5 if you are weighing a non-Claude option. For model rankings on coding benchmarks, see best LLM for coding.

What we tested

Nothing: we have not run these four models for this page. Every price, limit and positioning statement is Anthropic's, read from its docs and posts on 2026-10-26. The advice to move up or down is editorial judgement: the reliable way to choose is an eval on your own prompts and data, which is also Anthropic's advice.

  • Claude Code: Update now if you use the sandbox or run Claude Code on a cloud provider; skip if you wanted flashy new features, this is a hardening release. Tested (smoke test, 2026-10-25, local Qwen2.5-3B-Instruct, not vendor models).

FAQ

Which Claude model is best for coding?

Anthropic lists Opus 5.5 for complex agentic coding and Sonnet 5.5 for everyday coding. Anthropic's Haiku post says Sonnet 5.5 and Opus 5.5 remain the better choices for complex agentic coding. For benchmark numbers see our best LLM for coding page; we have not run the models ourselves.

Which Claude model is best for writing, research and planning?

Anthropic does not rank them by those tasks. Its guide lists content creation for Sonnet 5.5, knowledge work and complex enterprise work for Opus 5.5, and multistep deep research for Fable 5.1. Start with Opus 5.5 or Sonnet 5.5 and run your own prompts.

Is Haiku 5.5 good enough for subagents?

Anthropic says it pairs well with Opus 5.5 and Sonnet 5.5 as a subagent on coding work, and is best suited to narrowly scoped tasks such as compaction, summarization and subagent work.

Should I use Opus 5.5 or Fable 5.1?

Anthropic says to start with Opus 5.5 and move to Fable 5.1 if your evals at xhigh or max effort still fall short on demanding reasoning or long-horizon agentic work. Fable 5.1 costs $10/$50 against $4/$20.

How much cheaper is Sonnet 5.5 than Opus 5.5?

Half per token on list price: $2 input and $10 output against $4 and $20 per million tokens, per Anthropic's models overview on 2026-10-26. Real bills also depend on how many tokens each model uses for your task, so run your own prompts before assuming a saving.

Have you tested these Claude models?

No. We have not run Opus 5.5, Sonnet 5.5, Haiku 5.5 or Fable 5.1 for this page. Prices, limits and positioning are Anthropic's own statements, read on 2026-10-26, and the advice to move up or down is editorial judgement.

Sources

Page history

  • : First published. Anthropic's models overview, choosing-a-model guide, pricing page, Opus 5.5 and Haiku 5.5 posts and Claude Code model docs read on this date; nothing was run.

Markdown version for agents: /guides/which-claude-model-should-i-use.md