Claude Sonnet 5.5 and Haiku 5.5: What's New, Pricing, and Alternatives (2026)
Claude Sonnet 5.5 and Claude Haiku 5.5 are the two newest models in Anthropic's Claude 5.5 family. The Sonnet model launched on September 28, 2026, and Haiku 5.5 followed on October 7, 2026. Sonnet 5.5 costs $2 input and $10 output per million API tokens. Haiku 5.5 starts at $0.10 input and $0.50 output.
This guide covers what changed in each model and what the launch tests say about documents, slides, and spreadsheets. It also covers pricing, limits, and the alternatives worth testing. Every figure comes from vendor pages read on October 8, 2026.
What Anthropic released, and when
Three official channels confirm the dates. Anthropic's model pages show September 28 for the Sonnet release and October 7 for Haiku. The Claude Platform release notes carry the same two dates. The Claude Help Center release notes, which track the Claude apps, list both launches too.
The family now has three members. Claude Opus 5.5 arrived first, on September 22. Anthropic calls the new Sonnet "the second model in the Claude 5.5 family." It describes the model as "a faster, lower-cost complement to Claude Opus 5.5."
Haiku 5.5 completes the set. Anthropic introduces it as "the cheapest, fastest, and most capable small model we've ever released." That is the vendor's own claim, so treat it as a starting point for your tests.
For developers, the model IDs are claude-sonnet-5-5 and claude-haiku-5-5. Both models are available on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry, according to the platform release notes.
Claude Sonnet 5.5: what changed
Anthropic positions Sonnet 5.5 as the everyday model in the family. Its launch page says the model "is strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets." The same page adds: "It's also got a sharp eye for design."
Four changes stand out on the launch page:
- Speed. Anthropic says Sonnet 5.5 "generates outputs 30%+ faster than Sonnet 5." It calls it the fastest Sonnet model to date.
- Writing. Like Opus 5.5, the new Sonnet "writes more clearly than our previous generation of models," according to Anthropic.
- Long tasks and images. Anthropic says it is "the first Sonnet model to beat Pokémon Red working only from screenshots."
- Safety. It is the first Sonnet model to launch with cyber safeguards like those on Anthropic's most capable models.
Anthropic is also clear about the ceiling. The page says "Opus 5.5 remains clearly stronger at complex, open-ended work requiring sustained judgment." In practice, the Sonnet model handles the bulk of routine work, with Opus 5.5 kept for the hard cases.
One developer note matters before you switch. Teams that run Sonnet with thinking off need the new between_tools setting before moving to Sonnet 5.5. Anthropic's migration guide covers the details.
Claude Haiku 5.5: what changed
Haiku 5.5 targets volume. Anthropic says it "is designed for high-volume, cost-sensitive tasks." The launch page lists "summaries, compactions, database queries, and classification requests" as typical work.
The page also describes Haiku 5.5 as a helper for larger models. It "pairs well with Opus 5.5 and Sonnet 5.5 as a subagent on coding work." A footnote adds that Haiku 5.5 is Anthropic's fastest model at standard speed.
Haiku 5.5 is also "our first Haiku-class model to come with an adjustable effort setting." That lets you trade cost against quality per request, as you already can with the larger models.
Two customer quotes on the launch page describe report and deck work directly. Box says it would use Haiku 5.5 "on analytical work that runs at scale, from cost reports to financial summaries and weekly recurring reviews." Rogo describes a split job in its quote. "While a bigger model builds the deck, a Haiku 5.5 subagent goes into the 10-K and pulls the segment revenue line the deck needs."
That second pattern is worth copying. A large model plans and writes, and a small model fetches the numbers it needs.
What the launch tests say about documents, slides, and spreadsheets
Both launch pages include coding benchmarks. Three of the tests on these pages are more useful if your job ends in a report, a spreadsheet, or a deck.
GDPval-AA and AA-Briefcase: professional work
Anthropic describes GDPval-AA v2.1 as a test that "evaluates agents on real-world professional work across 44 occupations." The new Sonnet scores 1844 on it, two points below Opus 5.5 at 1846. Haiku 5.5 scores 1620.
On AA-Briefcase v1.1, a long-horizon knowledge work test, the Sonnet model scores 1811 and Haiku 5.5 scores 1578. Both tests were run by Artificial Analysis, and Anthropic notes a pre-release bug that may have slightly understated Sonnet 5.5's scores.
Chartography: reading charts
Chartography measures visual chart recognition. The new Sonnet scores 61.6% with no tools, and Haiku 5.5 scores 46.4%. Sonnet 5 scored 15.6% on the same table.
This is the test to watch if your inputs include chart screenshots, scanned reports, or slides from someone else's deck.
OSWorld: operating a computer
On OSWorld 2.1, the Sonnet model scores 80.1% on partial credit. Haiku 5.5 scores 72.4% on the offline subset. The two numbers use different scoring, so compare each model only with the rows Anthropic lists beside it.
Treat all of these as vendor-published results. Your own files, with your own formatting quirks, are the test that counts.
Pricing
The API prices below are the amounts shown on the Claude pricing page. Prices are per million tokens, and prompt caching prices reflect the 5-minute cache.
| Model | Input | Output | Cache read | Cache write |
|---|---|---|---|---|
| Claude Sonnet 5.5 | $2 | $10 | $0.10 | $2.50 |
| Claude Haiku 5.5, prompts up to 100K tokens | $0.10 | $0.50 | $0.01 | $0.125 |
| Claude Haiku 5.5, prompts over 100K tokens | $0.50 | $2.50 | $0.05 | $0.625 |
| Claude Opus 5.5 | $4 | $20 | $0.20 | $5 |
Haiku 5.5 is the only model in the table with two price bands. The band depends on prompt length, so long prompts cost more per token. Anthropic says prompts up to 100,000 tokens "make up around 90% of requests to our previous Haiku model."
Inside the Claude apps, cost works through plans rather than tokens. The pricing page lists Pro at $17 per month on the annual plan, or $20 billed monthly. Max starts from $100 per month.
Max and Team subscribers also get a new monthly API credit for the Claude Platform. The Haiku 5.5 page lists $100 per month for Max 5x, $200 for Max 20x, and up to $500 for Team, pooled across users.
Context window, output limits, and effort settings
Anthropic's models overview lists the same core limits for both models:
- Context window: 1M tokens.
- Maximum output: 128K tokens.
- Inputs: text and images, with text output.
The effort setting is the bigger practical change for Haiku users. The launch page charts Haiku 5.5 at five effort levels, from Low to Max, against cost per task. Anthropic's Sonnet 5.5 page explains the general pattern: higher effort usually means a higher cost per task and a higher score.
For recurring reports, that suggests a simple rule. Start at a low effort setting, check the output against a known-good report, and raise effort only where the low setting misses.
Sonnet 5.5, Haiku 5.5, or Opus 5.5: which to pick
Anthropic's own descriptions give a clear split:
- Use Opus 5.5 for long, high-stakes work that needs sustained judgment.
- Use Sonnet 5.5 for everyday documents, slides, and spreadsheets, and for most agent tasks.
- Use Haiku 5.5 for high-volume steps such as extraction, classification, summaries, and subagent lookups.
The models overview puts it in one line each. Sonnet 5.5 is "The best combination of speed and intelligence." Haiku 5.5 is "For high-volume, latency-sensitive tasks such as classification, extraction, and routing."
A reliable way to choose is to test on your own files. Run one real task through two models with the same inputs. Record the time, the cost per task, and every number you had to correct.
Alternatives worth knowing
Several other frontier releases landed in the same two weeks. The table lists the ones you can try today, plus one that is announced but not yet open.
| Model | Released | API price per million tokens | Where to use it |
|---|---|---|---|
| Claude Sonnet 5.5 | September 28, 2026 | $2 input / $10 output | Claude apps, Claude API, cloud platforms |
| Claude Haiku 5.5 | October 7, 2026 | From $0.10 input / $0.50 output | Claude API, cloud platforms |
| GPT-6.1 Sol | September 29, 2026 | $2 input / $10 output, up to 272K input tokens | ChatGPT Work, Codex, OpenAI API |
| GPT-6 Luna | September 22, 2026 | $0.10 input / $0.50 output, up to 272K input tokens | ChatGPT, OpenAI API |
| Gemini 4 Argon | September 30, 2026 | Not yet open to developers | Early testers only |
GPT-6.1 Sol. The ChatGPT release notes say it "improves on GPT-6 Sol in agentic coding, computer use, and professional work." It is rolling out in ChatGPT Work and Codex, starting with Pro users. OpenAI's API pricing page lists it at the same rates as Sonnet 5.5.
GPT-6 Luna. OpenAI's high-volume model is the closest match to Haiku 5.5 on price. The October 7 release notes say Free and Go users get GPT-6 Luna in the Chat tab, with that rollout starting the next day. Our GPT-6 Sol and Luna explainer covers its launch.
Gemini 4 Argon. Google DeepMind announced it on September 30. Its post says Google will gather feedback "before making Argon available to developers, enterprises, and consumers as soon as possible." For now, the post describes access for early testers only.
For the top of the Claude family, see our Claude Opus 5.5 write-up.
Where a faster model still needs a check
A lower price per token makes it easier to run more tasks. It does not remove the need to check the output. A report with one wrong figure is still a wrong report, whatever the model cost.
That check is where a workspace matters more than the model name. Powerdrill Bloom describes itself as "The AI data analyst that shows its work." Its homepage adds: "Every number comes back with the page, the row and the figure behind it."
The job shape matches the one Anthropic describes for its new Sonnet. You start with a spreadsheet or a set of documents and end with a finished file. The Powerdrill Bloom pricing page lists "Create slides, Office docs, Excel analysis, and Nano Banana images" on paid plans. Our Excel to PowerPoint page covers that route, and the AI report generator page covers written reports.
Pick the model by testing it. Then keep the source trail next to the output, so a reviewer can confirm each number in seconds.
How to start with Sonnet 5.5 and Haiku 5.5
If you use the Claude apps, pick Sonnet 5.5 for your next document, deck, or spreadsheet task. Give it one real file and one real deliverable. A quarterly summary makes a better test than a toy prompt.
If you build on the API, switch the model ID and rerun your evaluation set first. Compare cost per task, not just cost per token. A model that finishes in fewer steps can cost less overall.
If your pipeline has a high-volume step, try Haiku 5.5 there first. Extraction and first-pass summaries are good candidates, with a larger model reviewing the result.
When you want a spreadsheet turned into a sourced chart, report, or deck, you can try Powerdrill Bloom on your own files.
Frequently asked questions
What is Claude Sonnet 5.5?
It is the second model in Anthropic's Claude 5.5 family, released on September 28, 2026. Anthropic says it is strongest at everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets. API pricing is $2 input and $10 output per million tokens.
What is Claude Haiku 5.5 used for?
Anthropic designed it for high-volume, cost-sensitive tasks such as summaries, database queries, and classification. It also works as a subagent beside Opus 5.5 or Sonnet 5.5. It launched on October 7, 2026.
How much does Claude Haiku 5.5 cost?
For prompts up to 100K tokens, it costs $0.10 per million input tokens and $0.50 per million output tokens. For longer prompts, the rates are $0.50 and $2.50. Those are the amounts on the Claude pricing page.
What is the difference between Sonnet 5.5 and Opus 5.5?
Opus 5.5 costs $4 input and $20 output per million tokens, twice the Sonnet 5.5 rates. Anthropic says Opus 5.5 remains stronger at complex, open-ended work that needs sustained judgment. Sonnet 5.5 is the faster choice for well-scoped everyday work.
Why does Haiku 5.5 have two prices?
The price depends on prompt length. Prompts up to 100,000 tokens use the lower band, and longer prompts use the higher one. Anthropic says around 90% of requests to the previous Haiku model fell in the lower band.
Sources: Anthropic, Introducing Claude Sonnet 5.5 · Anthropic, Introducing Claude Haiku 5.5 · Claude Platform release notes · Claude Help Center release notes · Claude pricing · Claude models overview · ChatGPT release notes · OpenAI API pricing · Google DeepMind, Gemini 4 Argon. Prices and availability read from vendor pages on October 8, 2026.