THE AI RANKINGS

Anthropic

Claude Opus 5.5

Provider
Anthropic
Status
Current
Context
1,000,000 tok
Price
$4 / $20 /MTok
Knowledge
2026-06

Claude Opus 5.5 is Anthropic’s flagship model, released on 22 September 2026 as claude-opus-5-5 on the Claude API, in the Claude apps, and on Amazon Bedrock, Google Cloud and Microsoft Foundry (Anthropic). It succeeds Claude Opus 5 at $4/$20 per million tokens, 20% below Opus 5’s $5/$25, with cache reads down 60% to $0.20.

Artificial Analysis scored it 58 on its Intelligence Index v4.3 at max effort — the highest score it has recorded, five points clear of GPT-6 Astra and Claude Fable 5.1 on 53. It leads six of the index’s ten evaluations and trails on three. The model uses about 119,000 output tokens per index task against Opus 5’s 73,000, so the price cut leaves the cost per completed task roughly where it was: $5.98, against $5.86 for Opus 5. Anthropic published no SWE-bench figure for this release.

Quick specs

ProviderAnthropic
Released22 September 2026 (Claude API, Claude apps, Bedrock, Google Cloud, Microsoft Foundry)
API model IDclaude-opus-5-5
Context window1,000,000 tokens
Max output128,000 tokens (300K via the Batch API beta header)
Knowledge cutoffJune 2026
Input price$4.00 / MTok
Output price$20.00 / MTok
Cache reads$0.20 / MTok (5% of input)
Default effortMedium
ModalitiesText and image in; text out
AA Intelligence Index58 at max effort (highest recorded); 56 xhigh, 54 high, 51 medium
Cost per AA index task$5.98
LimitationsNo SWE-bench figure; 1.6x Opus 5’s output tokens per task; biology and cybersecurity requests re-route to Opus 4.8

TRY CLAUDE →

What’s new in Opus 5.5

Anthropic positions Opus 5.5 as the model to start with for most workloads, with Fable 5.1 reserved for demanding reasoning and long-horizon agentic work, or for cases where Opus 5.5 at higher effort still falls short. Three things changed from Opus 5:

The price. $4/$20 against $5/$25, with cache reads at $0.20 rather than $0.50 and cache writes at $5.00 rather than $6.25. Anthropic says typical workloads cost about 40% less.

The default effort. Opus 5.5 defaults to medium on the Claude API, where Opus 5 defaulted to high. On Artificial Analysis’s measurements the effort setting is worth seven index points — 51 at medium up to 58 at max — so the default is not the configuration behind the headline score.

The speed. Anthropic says output generation is more than 30% faster than Opus 5.

Benchmark performance

Anthropic’s launch table

BenchmarkClaude Opus 5.5Claude Fable 5.1
Terminal-Bench 4.066.4%55.8%
FrontierCode v1.154.4%50.3%
CursorBench 4.057.8%
GDPval-AA v2.11846 Elo
OSWorld 2.081.8%

Anthropic also says Opus 5.5 reaches parity with Fable 5.1 in biology and cybersecurity. These are vendor figures.

The independent read (Artificial Analysis)

MeasureClaude Opus 5.5Context
Intelligence Index v4.358 (max effort)Highest AA has measured; GPT-6 Astra and Fable 5.1 both 53
Index by effort51 medium, 54 high, 56 xhigh, 58 maxDefault effort is medium
Terminal-Bench 4.059.6%Level with GPT-6 Astra (xhigh); Anthropic’s own harness reports 66.4%
Output tokens per index task~119,000Opus 5: ~73,000
Cost per index task$5.98Opus 5: $5.86

Source: Artificial Analysis. It measured the model with Anthropic’s default fallback enabled. The Terminal-Bench gap between the two harnesses — 66.4% from Anthropic, 59.6% from Artificial Analysis — is the clearest illustration on this page of why vendor numbers are treated as a ceiling here.

The safeguard re-routing

Biology and cybersecurity requests that trigger Anthropic’s safeguards are served by Claude Opus 4.8 instead. Full capability on those tasks requires admission to Anthropic’s Life Sciences Verification Program or its expanded Cyber Verification Program. For anyone running agents across mixed workloads, the practical consequence is that a single run can be answered by more than one model, at different capability levels.

Pricing breakdown

Claude Opus 5.5Claude Opus 5
Input$4.00$5.00
Output$20.00$25.00
Cache reads$0.20$0.50
Cache writes$5.00$6.25
Cost per AA index task$5.98$5.86

The per-token cut does not carry through to the per-task cost, because the model spends about 1.6 times as many output tokens on the same work. The saving is real for workloads dominated by cached input, and close to nil for long agentic runs.

How Opus 5.5 compares

vs Claude Fable 5.1

Fable 5.1 costs $10/$50 against Opus 5.5’s $4/$20 and scores 53 on the index to Opus 5.5’s 58. Anthropic’s own table puts Opus 5.5 ahead on Terminal-Bench 4.0 (66.4% to 55.8%) and FrontierCode (54.4% to 50.3%), while its documentation still points to Fable 5.1 for demanding reasoning and long-horizon agentic work.

vs GPT-6 Astra

GPT-6 Astra scores 53 on the index at $10/$50 and $3.26 per task. Opus 5.5 is five index points ahead at less than half the token price, but costs more per completed task ($5.98), because it spends far more output tokens per task.

vs Claude Opus 5

Same context window and modalities, 20% lower prices, more than 30% faster output, and five index points higher at max effort (58 to 51 on the re-based scale). Cost per completed task is roughly unchanged.

Known limitations

No SWE-bench figure. Anthropic published no SWE-bench Pro or Verified result for this release, where Opus 5 reported 79.2% Pro and 96.0% Verified.

Token usage. About 1.6 times Opus 5’s output tokens per index task, which cancels most of the price cut on output-heavy work.

Benchmark gap between harnesses. Terminal-Bench 4.0 reads 66.4% on Anthropic’s harness and 59.6% on Artificial Analysis’s.

Re-routing on sensitive work. Biology and cybersecurity requests that trigger safeguards are answered by Opus 4.8 unless you are in one of Anthropic’s verification programmes.

Frequently asked questions

When was Claude Opus 5.5 released?

22 September 2026, as claude-opus-5-5 on the Claude API, in the Claude apps and on Amazon Bedrock, Google Cloud and Microsoft Foundry.

How much does Claude Opus 5.5 cost?

$4 per million input tokens and $20 per million output tokens, 20% below Claude Opus 5. Cache reads are $0.20 per million, 60% below Opus 5, and cache writes are $5.00. Batch requests are half price. Anthropic says typical workloads cost about 40% less than on Opus 5, though Artificial Analysis measured the cost of a completed Intelligence Index task at $5.98, against $5.86 for Opus 5, because the model uses more output tokens.

Is Claude Opus 5.5 the best AI model?

On the independent Artificial Analysis Intelligence Index it is the highest-scoring model measured to date: 58 at max effort, five points above GPT-6 Astra and Claude Fable 5.1. It leads six of the index’s ten evaluations and trails on three, and it publishes no SWE-bench figure. It is #1 on our best AI models ranking.

What is Claude Opus 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.

Why do some Claude Opus 5.5 requests get answered by Opus 4.8?

Biology and cybersecurity requests that trigger Anthropic’s safeguards are re-routed to Claude Opus 4.8. Full capability on those tasks requires Anthropic’s Life Sciences Verification Program or its expanded Cyber Verification Program.


Written at launch, 23 September 2026. Artificial Analysis figures are independent; the launch table rows are Anthropic’s. Pricing, specifications and availability are from Anthropic’s launch post and model documentation as of that date.