Claude Sonnet 5.5
- Provider
- Anthropic
- Status
- Current
- Context
- 1,000,000 tok
- Price
- $2 / $10 /MTok
- Knowledge
- 2026-06
Claude Sonnet 5.5 is Anthropic’s mid-tier model, released on 28 September 2026 as claude-sonnet-5-5 on the Claude API, in the Claude apps, and on Amazon Bedrock, Google Cloud and Microsoft Foundry (Anthropic). It succeeds Claude Sonnet 5 at the same $2/$10 per million tokens, and Anthropic says it generates output more than 30% faster and costs up to 30% less per task.
Artificial Analysis scored it 56 on its Intelligence Index at max effort — two points under Claude Opus 5.5 (58), and three above GPT-6 Astra and Claude Fable 5.1 (53). That is 18 points above Sonnet 5. It got there by using about 193,000 output tokens per index task, the most Artificial Analysis has measured, so a completed task cost $7.60 — more than Opus 5.5’s $5.98, despite half the token price. Anthropic published no SWE-bench figure for this release.
Quick specs
| Provider | Anthropic |
| Released | 28 September 2026 (Claude API, Claude apps, Bedrock, Google Cloud, Microsoft Foundry) |
| API model ID | claude-sonnet-5-5 |
| Context window | 1,000,000 tokens |
| Max output | 128,000 tokens (300K via the Batch API beta header) |
| Knowledge cutoff | June 2026 |
| Input price | $2.00 / MTok |
| Output price | $10.00 / MTok |
| Cache reads / writes | $0.20 / $2.50 per MTok |
| Default effort | High on the Claude API; medium in Claude Code and the Claude apps |
| Modalities | Text and image in; text out |
| AA Intelligence Index | 56 at max effort; 47 high, 41 medium, 36 low |
| Cost per AA index task | $7.60 |
| Limitations | No SWE-bench figure; the most output tokens per task AA has measured; high-risk cyber tasks fall back to Sonnet 5 |
What’s new in Sonnet 5.5
Anthropic positions Sonnet 5.5 as the faster, lower-cost complement to Opus 5.5: strongest on well-scoped everyday tasks, bug fixing, and producing documents, slides and spreadsheets, with Opus 5.5 kept for complex work that needs careful judgement. Its model documentation describes it as “the best combination of speed and intelligence” in the current lineup. Three things changed from Sonnet 5:
The capability jump. On Anthropic’s own table, Terminal-Bench 4.0 goes from 10.3% to 70.6% and OSWorld 2.1 from 57.0% to 80.1%, and the knowledge-work scores (GDPval-AA 1844, AA-Briefcase 1811) sit level with Opus 5.5. Artificial Analysis puts the gain at 18 index points.
The speed. Anthropic says output generation is more than 30% faster than Sonnet 5. Artificial Analysis measured about 142 output tokens per second.
The price, unchanged. $2/$10 per million tokens, with cache reads at $0.20 and cache writes at $2.50. The “up to 30% less per task” claim comes from efficiency — faster output and fewer tool calls, as VentureBeat reports it — not from a price cut. Anthropic also says Sonnet 5.5 is the first Sonnet model to finish Pokémon Red from screenshots alone.
A smaller Claude Haiku 5.5 is announced for “the coming weeks” and has not shipped.
Benchmark performance
Anthropic’s launch table
| Benchmark | Claude Sonnet 5.5 | Claude Sonnet 5 | Claude Opus 5.5 |
|---|---|---|---|
| Terminal-Bench 4.0 | 70.6% | 10.3% | 66.4% |
| FrontierCode 1.1 (Main) | 46.2% (max effort) | 42.4% | 54.4% |
| CursorBench 4.0 | 55.5% | 34.1% | 57.8% |
| GDPval-AA v2.1 | 1844 Elo | 1449 | 1846 |
| AA-Briefcase v1.1 | 1811 | 1359 | 1822 |
| Humanity’s Last Exam (with tools) | 64.5% | 54.9% | 67.7% |
| OSWorld 2.1 (partial) | 80.1% | 57.0% | 81.8% |
| Chartography (no tools) | 61.6% | 15.6% | 64.4% |
These are vendor figures. Anthropic’s table also prints GPT-6 Sol at 49.3% on FrontierCode, ahead of Sonnet 5.5’s 46.2%. Sonnet 5.5 beats Opus 5.5 on one row, Terminal-Bench 4.0, and trails it on the rest.
The independent read (Artificial Analysis)
| Measure | Claude Sonnet 5.5 | Context |
|---|---|---|
| Intelligence Index | 56 (max effort) | Second only to Opus 5.5 at max (58); GPT-6 Astra and Fable 5.1 both 53 |
| Index by effort | 36 low, 41 medium, 47 high, 56 max | API default is high |
| Terminal-Bench 4.0 | 64% | Opus 5.5 and GPT-6 Astra about 60%; Anthropic’s harness reports 70.6% |
| AA-Omniscience accuracy | 54% | Opus 5.5: 66% |
| Output tokens per index task | ~193,000 | The most AA has measured; about 60% more than Opus 5.5 or Sonnet 5 at max, ~7x GPT-6 Astra |
| Cost per index task | $7.60 | Opus 5.5: $5.98; about 50% more than Sonnet 5 |
| Output speed | ~142 tokens/s | — |
Source: Artificial Analysis and its model page. It measured the model with Anthropic’s default fallback enabled; Sonnet 5.5 fell back in about 0.1% of tasks, mostly to Sonnet 5. Artificial Analysis reports parity with Opus 5.5 on AA-Briefcase, GDPval-AA and AutomationBench-AA, and a gap of about six points to Opus 5.5 on Humanity’s Last Exam and SciCode.
The cost-per-task question
Anthropic’s launch claim and the independent measurement point in opposite directions, and both can be true. Anthropic says tasks cost up to 30% less than on Sonnet 5 in its own testing. Artificial Analysis, running its index at max effort, measured $7.60 per completed task — about 50% more than Sonnet 5, and more than Opus 5.5’s $5.98 — because the model spent about 193,000 output tokens per task.
The effort setting is the lever. The API default is high, where Artificial Analysis scores the model at 47 rather than 56. Buyers choosing Sonnet 5.5 over Opus 5.5 for cost should measure the per-task bill at the effort level they will run, not the per-token price.
Safeguards
High-risk cybersecurity tasks fall back to Claude Sonnet 5; Anthropic says routine debugging is unaffected, and cyber defenders can apply to its Cyber Verification Program for tiered access. Biology safeguards are the same as Sonnet 5’s. Anthropic also says new classifiers block attempts to extract the model’s reasoning for distillation.
Pricing breakdown
| Claude Sonnet 5.5 | Claude Sonnet 5 | Claude Opus 5.5 | |
|---|---|---|---|
| Input | $2.00 | $2.00 | $4.00 |
| Output | $10.00 | $10.00 | $20.00 |
| Cache reads | $0.20 | — | $0.20 |
| Cache writes | $2.50 | — | $5.00 |
| Batch (in / out) | $1.00 / $5.00 | $1.00 / $5.00 | 50% off |
| Cost per AA index task (max effort) | $7.60 | about $5 | $5.98 |
How Sonnet 5.5 compares
vs Claude Opus 5.5
Opus 5.5 scores two points higher on the index (58 to 56), leads on factual accuracy (66% to 54% on AA-Omniscience) and by about six points on Humanity’s Last Exam and SciCode, and costs less per completed index task ($5.98 to $7.60) at twice the token price. Sonnet 5.5 leads on Terminal-Bench 4.0 on both harnesses (70.6% to 66.4% on Anthropic’s; 64% to about 60% on Artificial Analysis’s) and is level on knowledge work.
vs GPT-6 Astra and Claude Fable 5.1
Both score 53 on the index, three under Sonnet 5.5, at $10/$50 per million tokens. GPT-6 Astra costs $3.26 per completed task and uses about a seventh of Sonnet 5.5’s output tokens; Fable 5.1 costs $7.63, about the same as Sonnet 5.5.
vs GPT-6 Sol
Same $2/$10 token price. GPT-6 Sol scores 48 on the index to Sonnet 5.5’s 56, but costs $1.06 per completed task against $7.60, and leads it on FrontierCode on Anthropic’s own table (49.3% to 46.2%).
vs Claude Sonnet 5
Same price, context window and modalities; 18 points higher on the index, more than 30% faster output on Anthropic’s figures, and a June 2026 knowledge cutoff against March. Sonnet 5 remains available as a legacy model on the Claude API.
Known limitations
No SWE-bench figure. Anthropic published no SWE-bench Pro or Verified result for this release, where Sonnet 5 reported 63.2% Pro and 85.2% Verified.
Token usage. About 193,000 output tokens per index task at max effort, the most Artificial Analysis has measured, which makes a completed task dearer than on Opus 5.5.
Factual accuracy. Artificial Analysis measured 54% on AA-Omniscience against Opus 5.5’s 66%.
Benchmark gap between harnesses. Terminal-Bench 4.0 reads 70.6% on Anthropic’s harness and 64% on Artificial Analysis’s.
Fallback on sensitive work. High-risk cybersecurity tasks are answered by Sonnet 5 unless you are in Anthropic’s Cyber Verification Program.
Frequently asked questions
When was Claude Sonnet 5.5 released?
28 September 2026, as claude-sonnet-5-5 on the Claude API, in the Claude apps and on Amazon Bedrock, Google Cloud and Microsoft Foundry.
How much does Claude Sonnet 5.5 cost?
$2 per million input tokens and $10 per million output tokens, the same as Claude Sonnet 5. Cache reads are $0.20 and cache writes $2.50 per million, and batch requests are half price. Anthropic says tasks cost up to 30% less than on Sonnet 5; Artificial Analysis measured $7.60 per completed Intelligence Index task at max effort, about 50% more than Sonnet 5, because the model uses more output tokens.
Is Claude Sonnet 5.5 better than Claude Opus 5.5?
Not overall. Opus 5.5 scores 58 on the Artificial Analysis Intelligence Index to Sonnet 5.5’s 56, is more accurate on factual questions, and costs less per completed index task at max effort. Sonnet 5.5 leads on Terminal-Bench 4.0 and is level on knowledge-work benchmarks, at half the token price. Opus 5.5 is #1 and Sonnet 5.5 #2 on our best AI models ranking.
What is Claude Sonnet 5.5’s context window?
1,000,000 tokens, with text and image input and text output, a 128,000-token maximum output (300,000 through the Batch API beta) and a June 2026 knowledge cutoff.
What is the default effort level for Claude Sonnet 5.5?
High on the Claude API, and medium in Claude Code and the Claude apps. Artificial Analysis scores the model 47 at high effort and 41 at medium, against 56 at max, so the default is not the configuration behind the headline score.
Written at launch, 29 September 2026. Artificial Analysis figures are independent; the launch table rows are Anthropic’s. Pricing, specifications and availability are from Anthropic’s launch page and model documentation as of that date.