Modelpeers
vs

Claude Opus vs Claude Sonnet: Is Opus worth twice the price of Sonnet?

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

Summary

Derived from sourced facts, not from builder opinions
  • This page compares Claude Opus 5.5 and Claude Sonnet 5.5, the current versions.
  • Sonnet 5.5 costs half as much per token: $2 input and $10 output per million, against $4 and $20 for Opus 5.5.
  • On coding they are close: 65.15% for Opus 5.5 and 64.14% for Sonnet 5.5 on Terminal-Bench 4.0, run by Vals AI in the Mini-SWE-agent harness. Opus cost less per test there ($13.20 against $16.51).
  • Sonnet 5.5 ranks first on the Vals Index (70.85%), ahead of Opus 5.5 (70.18%). Vals AI counted some Sonnet 5.5 and Opus 5.5 tasks that a fallback model answered after a refusal. Counting those as failures gives 69.03% and 64.16%.
  • Opus wins chat preference by a wide margin: rank 2 in Arena text against rank 28 for Sonnet.

Choose Claude Opus if

  • You run long agentic coding tasks, where Opus 5.5 cost less per test than Sonnet 5.5 in Vals AI's Terminal-Bench run.
  • You want the best Arena text and WebDev results.
  • You want Anthropic's default suggestion for most workloads.
  • You want shorter default answers: Opus 5.5 defaults to medium effort, Sonnet 5.5 to high.

Choose Claude Sonnet if

  • You pay per token and most requests are short.
  • You want a faster model (Anthropic rates Sonnet Fast, Opus Moderate).
  • You want to turn off up-front thinking with between_tools. Opus 5.5 always thinks.
  • You are on Claude Free, which includes Sonnet but not Opus.
  • You want half the per-token price.
Claude Opus
Claude Sonnet

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

Best by task

Agentic coding (Terminal-Bench 4.0 official board, both in Claude Code)

Tied

64.85% for Opus 5.5 against 61.82% for Sonnet 5.5, 330 trials each at max effort. The 95% confidence intervals overlap (plus or minus 3.11 and 2.94).

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

Tied

65.15% against 64.14%. A 1-point gap is too small to call.

Cost per test on Terminal-Bench 4.0 (Vals AI, API prices)

Claude Opus

$13.20 per test for Opus 5.5 against $16.51 for Sonnet 5.5, despite Opus's higher token price.

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

Claude Sonnet

Sonnet 5.5 70.85% against Opus 5.5 70.18%, and $18.70 per test against $28.12. Vals AI counted some Sonnet 5.5 and Opus 5.5 tasks that a fallback model answered after a refusal. Counting those as failures gives 69.03% and 64.16%.

Chat preference (Arena text)

Claude Opus

Rank 2 (1516) for Opus 5.5 (High) against rank 28 (1473) for Sonnet 5.5 (xHigh).

Web development (Arena WebDev)

Claude Opus

Rank 1 (1807) for Opus 5.5 (Max) against rank 3 (1767) for Sonnet 5.5 (xHigh), with confidence ranges that do not overlap.

Price per token

Claude Sonnet

$2 / $10 against $4 / $20 per million tokens.

Cost

One request in an agent loop that rereads a repository, so most input is cached. A full task makes many such requests. Illustrative, not measured.

Claude Opus

$0.124 per request

$124.00 a month for 1,000 requests

Claude Sonnet

$0.062 per request

$62.00 a month for 1,000 requests

Claude Sonnet costs 50% 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. It is not a measured task cost: models use different numbers of tokens, including reasoning tokens, for the same task.

Side by side

Claude OpusClaude Sonnet
Current versionClaude Opus 5.5 (claude-opus-5-5), released 2026-09-22
Claude Sonnet 5.5 (claude-sonnet-5-5), released 2026-09-28
API price per 1M tokens$4 input / $20 output · cache read $0.20 · Batch 50% off
$2 input / $10 output · cache read $0.10 · Batch 50% off
SpeedModerate (Anthropic's comparative latency)
Fast (Anthropic's comparative latency)
Reasoning settingsAdaptive thinking always on · default effort medium
Adaptive thinking · default effort high · up-front thinking can be turned off with between_tools
Context and output1M tokens context · 128K max output
1M tokens context · 128K max output
Anthropic's guidanceAnthropic suggests starting with Opus 5.5 for most workloads
Not published
In the Claude appNot on Free · included from Pro
Available on every plan, Free included
Breaking changes from Sonnet 5Not publishedFive breaking changes, including errors on forced tool use
Terminal-Bench 4.0 (official board)64.85%Opus 5.5 in Claude Code, max effort, rank 1, 330 trials61.82%Sonnet 5.5 in Claude Code, max effort, rank 2, 330 trials
Terminal-Bench 4.0 (Vals AI, same harness for all)65.15%Opus 5.5, rank 1, $13.20 per test64.14%Sonnet 5.5, rank 2, $16.51 per test
Vals Index v2.270.18%Opus 5.5, rank 2, $28.12 per test, includes tasks a fallback model answered after a refusal (64.16% if those count as failures)70.85%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)
Arena textRank 2 (1516)Opus 5.5 (High), 10,322 votesRank 28 (1473)Sonnet 5.5 (xHigh), 8,977 votes
Arena WebDevRank 1 (1807)Opus 5.5 (Max), 3,119 votesRank 3 (1767)Sonnet 5.5 (xHigh), 3,776 votes
Epoch Capabilities Index168.83Opus 5.5 (CI 165.25 to 173.95)166.18Sonnet 5.5 (CI 163.06 to 171.34)

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 Opus and Claude Sonnet

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.

Claude Opus

8 recorded incidents in the last 90 days.

Retiring

Claude Opus 4.8 fast mode is deprecated and ends November 30, 2026

Fast mode for Claude Opus 4.8 is deprecated and will be removed on November 30, 2026. After removal, requests with speed set to fast will return an error.

Takes effect Nov 30, 2026 · in 29 days

Who is affected:
Developers who send fast mode requests to Claude Opus 4.8 will get errors after removal.
Action needed:
Migrate fast mode requests to Claude Opus 5.5 before November 30, 2026.

Claude Sonnet

8 recorded incidents in the last 90 days.

FAQ

Is Claude Opus better than Sonnet?

For the current versions, Opus 5.5 leads Arena text (rank 2 against 28). On coding they are close: in Claude Code on the official Terminal-Bench 4.0 board, Opus 5.5 scored 64.85% and Sonnet 5.5 61.82%, within the confidence intervals. Sonnet 5.5 leads the Vals Index.

How much cheaper is Sonnet than Opus?

Sonnet 5.5 costs $2 input and $10 output per million tokens. Opus 5.5 costs $4 and $20, twice as much.

Is Opus more expensive per task?

Not always. On Vals AI's Terminal-Bench 4.0, Opus 5.5 cost $13.20 per test and Sonnet 5.5 $16.51.

Can I use Opus on the free Claude plan?

No. Free includes Sonnet and Haiku. Opus starts with Pro.

Which is faster, Opus or Sonnet?

Sonnet. Anthropic rates Sonnet 5.5 Fast and Opus 5.5 Moderate.

Do Opus and Sonnet have the same context window?

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

Which model does Anthropic suggest starting with?

Opus 5.5, for most workloads.

Related comparisons

Claude Sonnet vs Claude HaikuClaude Opus familyClaude Sonnet 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.