Claude Sonnet 5.5: Benchmarks, Pricing, and What Changed
Claude Sonnet 5.5 Guide
Claude Sonnet 5.5 is Anthropic’s new mid-tier model, released September 28, 2026. It keeps Sonnet 5’s $2/$10 price, generates output more than 30% faster, and comes within two points of Claude Opus 5.5 on several of Anthropic’s headline benchmarks — at half Opus 5.5’s price.
Quick answer: Claude Sonnet 5.5 (claude-sonnet-5-5) costs $2 input / $10 output per million tokens, the same as Sonnet 5, but Anthropic says it can cut the total cost of a task by up to 30% because it uses fewer tokens and tool calls. On Anthropic’s benchmarks it nearly matches Claude Opus 5.5 on knowledge work (1,844 vs 1,846 on GDPval-AA v2.1) and computer use (80.1% vs 81.8% on OSWorld 2.1), and edges it on Terminal-Bench 4.0 (70.6% vs 66.4%). Opus 5.5 keeps a clear lead on harder coding and reasoning tests.
What’s New in Claude Sonnet 5.5
Sonnet 5.5 is the second model in the Claude 5.5 family, six days after Opus 5.5. It replaces Sonnet 5 as Anthropic’s workhorse — the model built for “the best combination of speed and intelligence” — and Sonnet 5 moves to legacy status (still available until at least June 30, 2027).
- Faster: output generation is more than 30% faster than Sonnet 5.
- Cheaper per task, same price per token: up to 30% lower total cost on most work, because it needs fewer tokens and fewer tool calls. Slack measured about 14% fewer output tokens; Box saw tasks finish 2.4× faster with 12% fewer total tokens.
- A big capability jump over Sonnet 5: GDPval-AA v2.1 rises from 1,449 to 1,844 and OSWorld 2.1 from 57.0% to 80.1%.
- Newer knowledge: a June 2026 reliable knowledge cutoff, up from Sonnet 5’s January 2026.
- A thinking switch that fits speed-sensitive apps: a new
between_toolssetting turns off up-front thinking while keeping reasoning between tool calls.
Anthropic highlights document generation, summarization, spreadsheets, and bug fixing as natural fits. In customer testing, Base44 averaged 3.6 iterations per build where Opus 5 needed 7.7, and Zendesk processed tickets 20% faster.
Claude Sonnet 5.5 Specs and Pricing
| Property | Claude Sonnet 5.5 |
|---|---|
| API model ID | claude-sonnet-5-5 (Bedrock: anthropic.claude-sonnet-5-5) |
| Released | September 28, 2026 |
| Input / output price | $2 / $10 per million tokens — unchanged from Sonnet 5 |
| Prompt cache reads | $0.20 / MTok |
| Cache writes | $2.50 / MTok (5-minute) · $4 / MTok (1-hour) |
| Batch API | 50% off: $1 / $5 per MTok |
| Context window | 1M tokens (~555k words) — no 200K variant on the Claude API |
| Max output | 128K tokens (300K on the Batch API with a beta header) |
| Thinking | Adaptive, on by default; between_tools is the lowest setting |
| Default effort | high on the Claude API; medium in Claude Code and the Claude apps |
| Comparative latency | Fast |
| Reliable knowledge cutoff | June 2026 |
| Data retention | Zero data retention available on all platforms |
| Retirement | Not sooner than September 28, 2027 |
Price your own workload with the Claude API cost estimator. For how Sonnet pricing got here, see Sonnet 5 pricing: $2/$10 made permanent.
Claude Sonnet 5.5 Benchmarks
Anthropic’s published launch figures. Opus 5.5’s Terminal-Bench 4.0 score was run at xhigh effort.
| Benchmark | Sonnet 5.5 | Sonnet 5 | Opus 5.5 | GPT-6 Sol |
|---|---|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 70.6% | 10.3% | 66.4% | — |
| FrontierCode 1.1 Main (agentic coding) | 46.2% | 42.4% | 54.4% | 49.3% |
| CursorBench 4.0 (agentic coding) | 55.5% | 34.1% | 57.8% | — |
| GDPval-AA v2.1 (knowledge work, Elo) | 1,844 | 1,449 | 1,846 | 1,487 |
| AA-Briefcase v1.1 (knowledge work, Elo) | 1,811 | 1,359 | 1,822 | 1,483 |
| Humanity’s Last Exam (with tools) | 64.5% | 54.9% | 67.7% | — |
| OSWorld 2.1 (computer use, partial credit) | 80.1% | 57.0% | 81.8% | — |
| Chartography (chart recognition, no tools) | 61.6% | 15.6% | 64.4% | 53.6% |
Anthropic attaches caveats worth knowing. Sonnet 5.5’s FrontierCode score is from max effort, which scored lower than xhigh because max effort triggered a code-review routine that caused timeouts. Its GDPval-AA and AA-Briefcase runs hit a structured-outputs bug on a pre-release deployment, which Anthropic expects understates its real performance. The 10.3% for Sonnet 5 on Terminal-Bench 4.0 is Anthropic’s published figure.
For a row-by-row verdict on whether the half-price model is enough, see Claude Sonnet 5.5 vs Opus 5.5.
API Changes: Migrating From Sonnet 5
Five changes can break code that runs on Sonnet 5:
- Thinking can’t be switched fully off. If you run Sonnet 5 with thinking disabled, switch to
between_tools, which turns off up-front thinking and works athigheffort or below. - Forced tool use returns an error. Use
tool_choice: autowith strict tool use or structured outputs. - Thinking blocks are tied to the model and the conversation.
- The older
computer_20251124tool isn’t accepted on the Claude API and Google Cloud. - The advisor tool rejects Opus 4.8, Opus 4.7, and Sonnet 5 as advisor models.
As with Opus 5.5, text written between tool calls now arrives in thinking blocks. Set a display value that returns it, or use between_tools, if your UI shows progress updates. Sampling parameters (temperature, top_p, top_k) at non-default values return a 400 error.
Availability and Claude Code
- All platforms from launch: the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS, plus the Claude apps.
- Claude Code: the
sonnetalias now resolves to Sonnet 5.5 on the Anthropic API. It needs Claude Code v2.1.284 or later. Claude Code’s default model is Opus 5.5; switch with/model sonnetto save usage on everyday work. More in the Claude Code guide. - Safeguards: the first Sonnet with cybersecurity safeguards similar to Opus 5.5 — higher-risk cyber tasks fall back to Sonnet 5 — and the first with classifiers that block reasoning extraction. Biology safeguards match Sonnet 5.
Anthropic says Claude Haiku 5.5 will follow in the coming weeks, completing the 5.5 family. Until then, Haiku 4.5 remains the fastest option.
Limitations and Caveats
- Opus 5.5 still leads on the hardest coding and reasoning tests, by 8.2 points on FrontierCode and 3.2 on Humanity’s Last Exam.
- Five breaking API changes make migrating from Sonnet 5 more than a model-ID swap.
- Default effort differs by surface — high on the API, medium in Claude Code and the apps — so results can differ between them.
- Higher-risk cybersecurity requests are routed to Sonnet 5.
Frequently Asked Questions
When was Claude Sonnet 5.5 released?
Anthropic released Claude Sonnet 5.5 on September 28, 2026, six days after Claude Opus 5.5. Sonnet 5 becomes a legacy model and stays available until at least June 30, 2027.
How much does Claude Sonnet 5.5 cost?
The same as Sonnet 5: $2 per million input tokens and $10 per million output tokens, with cache reads at $0.20 and Batch API pricing of $1/$5. Anthropic says it can still cut the total cost of a task by up to 30% because it uses fewer tokens and tool calls.
Is Claude Sonnet 5.5 as good as Opus 5.5?
Close on some tasks, not all. On Anthropic’s benchmarks Sonnet 5.5 nearly ties Opus 5.5 on GDPval-AA v2.1 (1,844 vs 1,846) and OSWorld 2.1 (80.1% vs 81.8%), and scores higher on Terminal-Bench 4.0 (70.6% vs 66.4%). Opus 5.5 leads on FrontierCode (54.4% vs 46.2%), CursorBench 4.0, and Humanity’s Last Exam. Sonnet 5.5 costs half as much.
How much faster is Claude Sonnet 5.5 than Sonnet 5?
Anthropic says Sonnet 5.5 generates output more than 30% faster than Sonnet 5. Customer tests reported similar or larger gains, such as Box measuring tasks 2.4 times faster and Zendesk processing tickets 20% faster.
Can I turn off thinking on Claude Sonnet 5.5?
Not entirely. The lowest setting is between_tools, which turns off up-front thinking while keeping reasoning between tool calls; it works at high effort or below. Code that ran Sonnet 5 with thinking disabled needs to switch to between_tools.
How do I use Claude Sonnet 5.5 in Claude Code?
Update to Claude Code v2.1.284 or later with claude update, then run /model sonnet — on the Anthropic API the sonnet alias resolves to Sonnet 5.5. Claude Code’s default model is Opus 5.5, so switching to Sonnet 5.5 is a way to stretch usage limits on everyday work.
Is Claude Haiku 5.5 coming?
Anthropic says Claude Haiku 5.5 will be released in the coming weeks as the final model in the 5.5 family. Until then, Claude Haiku 4.5 is the fastest and cheapest current Claude model at $1/$5 per million tokens.