Modelpeers
vs

Claude Sonnet vs Claude Haiku: When is the cheaper model good enough?

Facts checked Oct 22, 2026. Every number links to its source.

Summary

Derived from sourced facts, not from builder opinions
  • This page compares Claude Sonnet 5.5 and Claude Haiku 5.5, the current versions.
  • Haiku 5.5 costs 20 times less per token for prompts up to 100K tokens: $0.10 input and $0.50 output per million, against $2 and $10 for Sonnet 5.5.
  • Sonnet 5.5 is much stronger on hard tasks: 64.14% against 36.36% on Terminal-Bench 4.0 run by Vals AI in the Mini-SWE-agent harness, and 70.85% against 58.71% on the Vals Index.
  • Anthropic positions Haiku for classification, routing, extraction and subagent tasks.

Choose Claude Sonnet if

  • The task is open-ended coding, research or writing.
  • Quality per answer matters more than cost per answer.
  • Your prompts often run over 100K tokens, where Haiku's per-token price rises five times.

Choose Claude Haiku if

  • You run high volumes of classification, extraction or routing.
  • You need the fastest responses.
  • You run many subagents under a stronger planner model.
  • You want about a twentieth of Sonnet's per-token price on prompts up to 100K tokens.
Claude Sonnet
Claude Haiku

Follow both to get an email when either one changes its price, limits or behavior.

Best by task

Agentic coding (Terminal-Bench 4.0, Vals AI, Mini-SWE-agent harness)

Claude Sonnet

64.14% against 36.36%.

Work tasks across finance, coding, legal and tax (Vals Index)

Claude Sonnet

70.85% (rank 1) against 58.71% (rank 11). Counting tasks a fallback model answered after a refusal as failures, Sonnet 5.5 still scores 69.03%.

Cost per test on the Vals Index (API prices)

Claude Haiku

$3.17 per test against $18.70.

Chat preference (Arena text)

Claude Sonnet

Rank 28 (1473) against rank 153 (1401).

Web development (Arena WebDev)

Claude Sonnet

Rank 3 (1767) against rank 35 (1585).

Speed

Claude Haiku

Anthropic rates Haiku 5.5 Fastest and Sonnet 5.5 Fast.

Cost

A short ticket plus a cached instruction block, answered with one label. Illustrative, not measured.

Claude Sonnet

$0.0037 per request

$3.70 a month for 1,000 requests

Claude Haiku

$0.0002 per request

$0.195 a month for 1,000 requests

Claude Haiku costs 95% less for this workload at list prices.

Estimate per request: (input × input rate + cached input × cached rate + output × output rate) ÷ 1,000,000, at list prices on the cited pricing pages. For Claude Haiku, a request with more than 100,000 input tokens is charged 5 times the input rates and 5 times the output rate. It is not a measured task cost: models use different numbers of tokens, including reasoning tokens, for the same task.

Side by side

Claude SonnetClaude Haiku
Current versionClaude Sonnet 5.5 (claude-sonnet-5-5), released 2026-09-28
Claude Haiku 5.5 (claude-haiku-5-5), released 2026-10-07
API price per 1M tokens$2 input / $10 output · cache read $0.10 · Batch 50% off
$0.10 input / $0.50 output for prompts up to 100K tokens · $0.50 / $2.50 above 100K · cache read $0.01 · Batch 50% off
SpeedFast (Anthropic's comparative latency)
Fastest (Anthropic's comparative latency)
Reasoning settingsAdaptive thinking · default effort high · up-front thinking can be turned off with between_tools
Adaptive thinking · default effort medium
Context and output1M tokens context · 128K max output
1M tokens context · 128K max output
Breaking changes from Sonnet 5Five breaking changes, including errors on forced tool use
Not published
In the Claude appAvailable on every plan, Free included
Available on every plan, Free included
Built forNot publishedClassification, routing, extraction and subagent tasks
Token countNot publishedSame text counts about 30% more tokens than on Haiku 4.5 (newer tokenizer)
Terminal-Bench 4.0 (Vals AI, same harness for all)64.14%Sonnet 5.5, rank 2, $16.51 per test36.36%Haiku 5.5, rank 16, $3.12 per test
Vals Index v2.270.85%Sonnet 5.5, rank 1, $18.70 per test, includes tasks a fallback model answered after a refusal (69.03% if those count as failures)58.71%Haiku 5.5, rank 11, $3.17 per test
Arena textRank 28 (1473)Sonnet 5.5 (xHigh), 8,977 votesRank 153 (1401)Haiku 5.5 (High), 6,080 votes
Arena WebDevRank 3 (1767)Sonnet 5.5 (xHigh), 3,776 votesRank 35 (1585)Haiku 5.5 (High), 3,860 votes
Epoch Capabilities Index166.18Sonnet 5.5 (CI 163.06 to 171.34)159.32Haiku 5.5 (CI 156.19 to 162.47)

Scores appear side by side only when both were published for the same test. Settings differ between sources, so small gaps can reverse under another harness.

Updates

Both Claude Sonnet and Claude Haiku

Now available

Full Claude Platform available on Google Cloud (Anthropic managed)

Claude Models on Google Cloud (Anthropic managed) is generally available with US and Global endpoints. It gives Google Cloud customers the full Claude Platform feature set with Google Cloud IAM, billing and commitment drawdown. Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 are available, and fast mode for Opus 5.5 costs $8 / $40 per million input and output tokens there.

Who is affected:
Teams that buy Claude through Google Cloud and want features that the Google-managed offering lacks.
What to do:
No action needed. Google Cloud customers who need same-day Claude API features can switch to the Anthropic-managed offering.

Breaking change

New API accounts can no longer edit context around thinking blocks

To counter distillation, the API now checks that a thinking block is sent back with the same system prompt, tools and messages. On Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, accounts created on or after August 31, 2026 get an error unless they opt into dropping the block.

Who is affected:
Agents that compact or rewrite earlier turns, especially on new API accounts.
Action needed:
Use server-side compaction or mid-conversation tool changes, or opt into non-strict mode.

Claude Sonnet

8 recorded incidents in the last 90 days.

Claude Haiku

2 recorded incidents in the last 90 days.

New release

Claude Haiku 5.5 released at a much lower price than Haiku 4.5

Claude Haiku 5.5 launched with a 1M token context window at $0.10 / $0.50 per million tokens for prompts up to 100,000 tokens and $0.50 / $2.50 above that. Haiku 4.5 costs $1 / $5.

Before

Claude Haiku 4.5: $1 / $5 per MTok

After

Claude Haiku 5.5: $0.10 / $0.50 per MTok (prompts up to 100k tokens)

Who is affected:
Apps calling claude-haiku-4-5 for high-volume work.
What to do:
Test claude-haiku-5-5. Remove budget_tokens, sampling parameters and prefill first.

FAQ

Is Claude Haiku good enough?

For classification, routing and extraction, Anthropic positions Haiku 5.5 for exactly that work. For coding it scored 36.36% on Terminal-Bench 4.0 run by Vals AI, against 64.14% for Sonnet 5.5.

How much cheaper is Haiku than Sonnet?

Haiku 5.5 costs $0.10 input and $0.50 output per million tokens for prompts up to 100K tokens. Sonnet 5.5 costs $2 and $10.

Is Haiku faster than Sonnet?

Yes. Anthropic rates Haiku 5.5 Fastest and Sonnet 5.5 Fast.

Are Sonnet and Haiku on the free Claude plan?

Yes. Claude Free includes Sonnet and Haiku. Opus starts with Pro.

Is Haiku more expensive for long prompts?

Yes. Above 100K tokens, Haiku 5.5 costs $0.50 input and $2.50 output per million tokens, five times its rate for shorter prompts.

Do Sonnet and Haiku have the same context window?

Yes. Both current versions take 1M tokens with 128K of output.

What is Claude Haiku best for?

Anthropic built Haiku 5.5 for classification, routing, extraction and subagent tasks.

Related comparisons

Claude Opus vs Claude SonnetClaude Sonnet familyClaude Haiku family

Sources and method

Prices, specs and limits come from each maker's own pages. Benchmark scores come from independent benchmark publishers and the makers' announcements, with the setting each was measured in. Changes come from changelogs, deprecation pages and status pages. The summary is written from these facts only.