Download

Claude Opus 5.5 vs Claude Sonnet 5.5

Claude Sonnet 5.5 is the better default for agent coding: it costs half as much as Claude Opus 5.5 per token and scored higher on Terminal-Bench 4.0, 70.6% against 66.4%. Opus 5.5 is ahead on FrontierCode 1.1 by 8.2 points, so it earns its price on the hardest engineering tasks.

How Opus 5.5 and Sonnet 5.5 differ

On paper, only in price. Both read up to 1M tokens, both write up to 128K tokens in one reply, and both take text and images. Anthropic released them six days apart, Opus 5.5 on 22 September 2026 and Sonnet 5.5 on 28 September. Opus costs $4 input and $20 output per 1M tokens, Sonnet $2 and $10, and cached input is $0.20 against $0.10.

The benchmarks split. Terminal-Bench 4.0 runs agentic coding tasks in a real terminal, and Sonnet won it, 70.6% to 66.4%, even though Anthropic ran Opus at xhigh effort. FrontierCode 1.1, a set of hard software engineering tasks, went to Opus, 54.4% to 46.2%, both at max effort. On Humanity's Last Exam with tools Opus leads by 3.2 points.

Which one to run in a coding agent

Our feature session, 60 calls with 80K-token prompts and 2K-token replies, costs $21.60 on Opus 5.5 and $10.80 on Sonnet 5.5 at list prices. With 90% of each prompt read from the cache it is $5.18 against $2.59, and a long agent day comes to $195 against $97.50. Every figure doubles on Opus because every rate does.

We start a task on Sonnet 5.5 and move it to Opus 5.5 when the work is large, vague or spread across many files, which is closer to what FrontierCode tests. Opus vs Sonnet for coding sets out when the dearer model pays for itself, and changing the model in Claude Code shows how to switch mid-session.

What these numbers leave out

The scores are Anthropic's own runs at the effort it chose, and the prices are API list prices in US dollars. A Claude plan bills by the month, not by the token, so on a plan the gap shows up in how far your limits stretch rather than on an invoice.

Side by side

Claude Opus 5.5 and Claude Sonnet 5.5: specifications and prices
Claude Opus 5.5Claude Sonnet 5.5
MakerAnthropicAnthropic
Released22 September 2026[1]28 September 2026[2]
Context window1M tokens[1]1M tokens[2]
Max output128K tokens[1]128K tokens[2]
Input price$4 per 1M tokens[1]$2 per 1M tokens[2]
Output price$20 per 1M tokens[1]$10 per 1M tokens[2]
Cached input$0.20 per 1M tokens[3]$0.10 per 1M tokens[3]
Long promptsOne priceOne price
Price changeNone announcedNone announced
Open weightsNo[1] Offered through the Claude API and cloud platforms; no weights are published.No[2] Offered through the Claude API and cloud platforms; no weights are published.
Readstext, image[1]text, image[2]

Benchmarks

Scores as the makers published them, each with its source and date. Makers run these at different settings, so read a gap of a few points as a tie.

Published benchmark scores (higher is better)
Claude Opus 5.5Claude Sonnet 5.5
Terminal-Bench 4.0Claude Opus 5.5: 66.4%66.4%Claude Sonnet 5.5: 70.6%70.6%FrontierCode 1.1 (Main)Claude Opus 5.5: 54.4%54.4%Claude Sonnet 5.5: 46.2%46.2%Humanity's Last Exam (with tools)Claude Opus 5.5: 67.7%67.7%Claude Sonnet 5.5: 64.5%64.5%
Benchmarks: Claude Opus 5.5 and Claude Sonnet 5.5
BenchmarkClaude Opus 5.5Claude Sonnet 5.5
Terminal-Bench 4.0Agentic coding tasks in a real terminal.66.4%[4] Run at xhigh effort.70.6%[5]
FrontierCode 1.1 (Main)Hard software engineering tasks.54.4%[4] Max effort.46.2%[5] Max effort.
Humanity's Last Exam (with tools)Expert-level questions across many fields, tools allowed.67.7%[4]64.5%[5]

What a coding session costs

Our own estimate from the list prices above, for three sessions we define: a coding agent resends the conversation with every call, so prompts are large and grow. "Cached" assumes 90% of prompt tokens are read from the prompt cache and leaves out cache write fees. A long-prompt price applies when a call's prompt is over its threshold. Work out your own mix with the token cost calculator.

Feature session at list prices, US dollars (lower is cheaper)
Claude Opus 5.5Claude Sonnet 5.5
No cachingClaude Opus 5.5: $21.60$21.60Claude Sonnet 5.5: $10.80$10.8090% of prompt tokens from cacheClaude Opus 5.5: $5.18$5.18Claude Sonnet 5.5: $2.59$2.59
Session costs: Claude Opus 5.5 and Claude Sonnet 5.5
SessionClaude Opus 5.5Claude Opus 5.5 cachedClaude Sonnet 5.5Claude Sonnet 5.5 cached
Quick fix10 calls of 25K prompt and 1.5K output tokens$1.30$0.45$0.65$0.22
Feature session60 calls of 80K prompt and 2K output tokens$21.60$5.18$10.80$2.59
Long agent day300 calls of 150K prompt and 2.5K output tokens$195$41.10$97.50$20.55

Value for money and release gap

Claude Sonnet 5.5 came out 6 days later. Price per point divides our feature-session cost by a published score: lower means more score for the money. It is a rough lens, not a ranking.

Value for money: Claude Opus 5.5 and Claude Sonnet 5.5
MeasureClaude Opus 5.5Claude Sonnet 5.5
Input tokens per $1250K500K
Filling the whole context window once$4.00$2.00
Feature session per point of Terminal-Bench 4.0$0.33$0.15
Feature session per point of FrontierCode 1.1 (Main)$0.40$0.23
Feature session per point of Humanity's Last Exam (with tools)$0.32$0.17

Questions

Is Claude Sonnet 5.5 better than Opus 5.5 for coding?

On Terminal-Bench 4.0, yes: 70.6% against 66.4%. On FrontierCode 1.1, no: Opus 5.5 scored 54.4% and Sonnet 5.5 46.2%. Sonnet costs half as much, so it is the sensible default, with Opus kept for the hardest tasks.

How much more does Opus 5.5 cost than Sonnet 5.5?

Twice as much on every rate: $4 against $2 per 1M input tokens, $20 against $10 per 1M output tokens and $0.20 against $0.10 for cached input. Our feature session costs $21.60 on Opus and $10.80 on Sonnet.

Do Opus 5.5 and Sonnet 5.5 have the same context window?

Yes. Both read up to 1M tokens and write up to 128K tokens in one reply, so the choice between them comes down to price and the task.

Most compared

Any other two models: pick them in the lists at the top of the page.

More tools

Sources

  1. platform.claude.com: docs/en/models/opus-5-5/overview, read
  2. platform.claude.com: docs/en/models/sonnet-5-5/overview, read
  3. platform.claude.com: docs/en/about-claude/pricing, read
  4. anthropic.com: claude-opus-5-5, read
  5. anthropic.com: claude-sonnet-5-5, read

Numbers from the makers' own pages, checked by hand, last read . List prices in US dollars before tax. Spotted a mistake? office@shuriken.sh.

Made by Shuriken, the harness for coding agents. Its hot swapping moves your chats to another of your accounts or API keys when one nears its limit. Use only accounts you own and are entitled to use.

Try Shuriken on your own computer

Add your Claude, Codex and Gemini accounts once. Shuriken switches between them so your agents keep working.

Download Pricing
Download on the App StoreGet it on Google PlayComing soon