How good is Claude Sonnet 5.5 at coding?

Vendor-reported scores as of September 28, 2026: 70.6% on Terminal-Bench 4.0 versus 10.3% for Sonnet 5 and 66.4% for Opus 5.5 at xhigh; 52.1% on FrontierCode 1.1 at xhigh versus 42.4% for Sonnet 5, 54.4% for Opus 5.5 and 49.3% for GPT-6 Sol; and 55.5% on CursorBench 4.0 versus 34.1% for Sonnet 5 and 57.8% for Opus 5.5. At High effort on FrontierCode it scores 10 points above Sonnet 5 at about one fifteenth of the cost per task. Anthropic says Opus 5.5 remains clearly stronger on complex, open-ended work.

How much does Claude Sonnet 5.5 cost compared to Opus 5.5?

Sonnet 5.5 is exactly half the price of Opus 5.5 on every line: $2 vs $4 per million input tokens, $10 vs $20 per million output tokens, $2.50 vs $5 per million for cache writes. Cache reads are $0.20 on both. Because Sonnet 5.5 also batches tool calls and uses fewer tokens than Sonnet 5, the per-task gap is often larger than the list-price gap: Balyasny Asset Management measured about 121K tokens per answer on Sonnet 5.5 versus 497K on Sonnet 5.

Where is Claude Sonnet 5.5 available?

As of September 28, 2026, Claude Sonnet 5.5 is generally available on the Claude Platform (model ID claude-sonnet-5-5), Amazon Web Services, Google Cloud and Microsoft Azure, with zero-data-retention support. It is the default effort Medium in Claude Code and the Claude apps and High on the Claude Platform. Claude Haiku 5.5, the high-volume tier, is due 'in the coming weeks'.

Quick Answer

What Is Claude Sonnet 5.5? Anthropic's New Model (Sep 2026)

Q: What is Claude Sonnet 5.5?

Claude Sonnet 5.5 is Anthropic's mid-tier model, released September 28, 2026 as the second member of the Claude 5.5 family after Opus 5.5 (September 22). It is priced identically to Sonnet 5 at $2 per million input tokens and $10 per million output tokens (cache reads $0.20, cache writes $2.50), generates output 30%+ faster, and needs fewer tokens per task, which Anthropic says cuts cost per task by up to 30%. It has a 1M-token context window, 128K max output, a June 2026 knowledge cutoff, adaptive thinking on by default and five effort levels (low, medium, high, xhigh, max). The API model ID is claude-sonnet-5-5.

Published: September 29, 2026

The short answer

Claude Sonnet 5.5 is Anthropic’s new mid-tier model, released September 28, 2026, at the same $2/$10 per million tokens as Sonnet 5 but with a step-change in agentic coding (70.6% on Terminal-Bench 4.0, up from 10.3%), 30%+ faster output and up to 30% lower cost per task. It is the second model in the Claude 5.5 family, six days behind Claude Opus 5.5, and it is positioned as the faster, cheaper complement: well-scoped everyday tasks, bug fixing, and polished documents, slides and spreadsheets. Claude Haiku 5.5 is next, “in the coming weeks.”

Specs at a glance

	Claude Sonnet 5.5	Claude Sonnet 5	Claude Opus 5.5
Released	Sep 28, 2026	Jun 30, 2026	Sep 22, 2026
Input / output (per MTok)	$2 / $10	$2 / $10	$4 / $20
Cache read / 5m write	$0.20 / $2.50	$0.20 / $2.50	$0.20 / $5
Context / max output	1M / 128K	1M / 128K	1M / 128K
Knowledge cutoff	June 2026	—	June 2026
Thinking	Adaptive, on by default	Adaptive	Adaptive, always on
Effort levels	low, medium, high, xhigh, max	4	5
Default effort	High (API), Medium (apps, Claude Code)	High	Medium
Output speed vs Sonnet 5	30%+ faster	—	—
Min cacheable prompt	512 tokens	1,024	512
Model ID	`claude-sonnet-5-5`	`claude-sonnet-5`	`claude-opus-5-5`
Clouds	Claude Platform, AWS, Google Cloud, Azure	same	same

The pricing line matters more than it looks: Sonnet 5.5 is now exactly half of Opus 5.5 on input, output and cache writes, and it matches GPT-6 Sol’s $2/$10 (see Sonnet 5.5 vs GPT-6 Sol vs Grok 4.7).

Benchmarks (vendor-reported, September 28, 2026)

Benchmark	Sonnet 5.5	Sonnet 5	Opus 5.5	GPT-6 Sol
Terminal-Bench 4.0 (agentic coding)	70.6%	10.3%	66.4% (xhigh)	not reported
FrontierCode 1.1 (Main)	52.1% xhigh / 46.2% max	42.4%	54.4%	49.3%
CursorBench 4.0	55.5%	34.1%	57.8%	—
GDPval-AA v2.1 (knowledge work)	1844	1449	1846	1487
AA-Briefcase v1.1	1811	1359	1822	1483
Humanity’s Last Exam (with tools)	64.5%	54.9%	67.7%	—
OSWorld 2.1 (computer use, partial)	80.1%	57.0%	81.8%	—
Chartography (no tools)	61.6%	15.6%	64.4%	53.6%

Three things stand out. First, the Terminal-Bench jump from 10.3% to 70.6% is the largest single-generation move Anthropic has published for a Sonnet, and it puts Sonnet 5.5 above Opus 5.5 on that one test. Second, on GDPval-AA (real tasks across 44 occupations) Sonnet 5.5 is two points behind Opus 5.5 and about 400 points above Sonnet 5. Third, the FrontierCode “max” score (46.2%) is lower than “xhigh” (52.1%): at max effort the model more often spun up multi-agent code review, which caused timeouts and out-of-scope edits that FrontierCode penalizes. Anthropic’s own framing is that Opus 5.5 “remains clearly stronger at complex, open-ended work requiring sustained judgment.” GPT-6 Sol numbers are Anthropic’s runs; OpenAI did not publish Terminal-Bench for Sol.

What changed versus Sonnet 5

Speed and efficiency. 30%+ faster generation, and fewer tokens and tool calls per task. Early testers saw it batch tool calls more aggressively than Sonnet 5. Customer data in the launch post: Balyasny Asset Management ~121K tokens per answer versus 497K on Sonnet 5; Base44 3.6 iterations per app build where Opus 5 took 7.7; Slack ~14% fewer output tokens with better results on almost all Slackbot evals; Zendesk 20% faster ticket handling.
Vision and long-horizon work. First Sonnet to beat Pokémon Red from screenshots alone; Chartography jumps from 15.6% to 61.6%.
Writing and design. Clearer prose than the previous generation; testers noted it follows slide templates well enough that a 10-slide earnings review draft was judged “ready to send as is” by two experts in an internal test.
API surface. thinking: {"type": "disabled"} is gone (use between_tools), forced tool_choice of any/tool returns a 400, thinking blocks are signed over the conversation for accounts created on or after August 31, 2026, computer use moves to computer_toolset_20260801 on the Claude API and Google Cloud, and the minimum cacheable prompt drops to 512 tokens. Full checklist in how to migrate to Claude Sonnet 5.5.

Safety and safeguards

Because Sonnet 5.5’s cyber capabilities are comparable to Opus 5’s, it is the first Sonnet to ship with real-time cyber safeguards: routine bug finding and fixing works, but higher-risk cybersecurity requests visibly fall back to Sonnet 5. Anthropic’s expanded Cyber Verification Program will offer tiered access on Sonnet 5.5, Opus 5.5 and Mythos models. Biology safeguards are unchanged from Sonnet 5. It is also the first Sonnet with reasoning-extraction classifiers (to blunt distillation attacks, the subject of Anthropic’s September 2026 threat report), and it expands preserved thinking so that thinking blocks cannot be moved between accounts. On Anthropic’s ~1,850-scenario behavioral audit it matches or improves on Sonnet 5 on most alignment measures, and on containment evaluations it is the least likely of any Claude model to probe the limits of its sandbox. A refusal now returns stop_reason: "refusal" with a category (cyber, bio, frontier_llm, reasoning_extraction, general_harms).

Who should use it

Agentic coding on a budget: Sonnet 5.5 at Medium effort beats Sonnet 5’s best Terminal-Bench score for less than a tenth of the cost per task. Start at medium for well-specified tasks and high for longer ones.
Documents, slides, spreadsheets, UI polish: this is where Anthropic explicitly points it.
Chat and latency-sensitive products: low or medium effort; 30%+ faster than Sonnet 5.
Not for: open-ended, multi-hour judgment work — that is still Opus 5.5 territory. Decision rules in Sonnet 5.5 vs Opus 5.5.

Last verified: September 29, 2026. All benchmark figures are Anthropic’s own; no independent Artificial Analysis index score for Sonnet 5.5 was published as of this date.