When was Claude Opus 5.5 released and what are the main changes?

Anthropic released Claude Opus 5.5 on September 22, 2026, introducing a significant pricing overhaul that reduces operational costs by approximately 40% compared to its predecessor. This update delivers performance metrics rivaling the larger 'Fable' tier models while maintaining stricter safety guardrails aligned with CEO Dario Amodei's philosophy.

How much does Claude Opus 5.5 cost compared to Opus 5?

The new pricing structure sets output tokens at $20 per million, down from $25, and input tokens at $4 per million, a reduction from the previous $5 rate. Combined with optimized compute requirements that make responses more than 30% faster, these adjustments result in roughly 40% lower costs for typical workloads.

How does Claude Opus 5.5 perform against competitors and previous models?

On the GDPval-AA v2.1 benchmark testing real-world work across 44 occupations, Opus 5.5 scored 1,846 points, surpassing both its predecessor and Anthropic’s own Fable 5.1 model. In coding-specific tests like Terminal-Bench 4.0, it achieved a score of 66.4%, exceeding OpenAI’s GPT-6 Astra which scored 57.9%.

What safety measures are included in Claude Opus 5.5?

Opus 5.5 implements strict safeguards similar to those in the restricted 'Mythos' tier, limiting its use for discovering exploits or developing biological weapons due to its high capabilities in biology and cybersecurity. Extended thinking features are permanently enabled to ensure consistent safety monitoring, and the model achieved the highest scores on Anthropic’s automated behavioral audit to date.

Where can I access Claude Opus 5.5?

The model is available immediately via the Anthropic API on major cloud providers including AWS, Google Cloud, and Microsoft Azure. This release serves as the first entry in the Claude 5.5 family, with Sonnet 5.5 and Haiku 5.5 scheduled for release in the coming weeks.

Anthropic Opus 5.5: Fable Performance at Lower Cost

TL;DR: Anthropic released Claude Opus 5.5, which delivers Fable-tier performance at a 40% lower cost than its predecessor. This strategic move aligns with CEO Dario Amodei’s ‘pace the frontier’ philosophy by prioritizing safety and efficiency over raw capability scaling.

Key facts

Anthropic released Claude Opus 5.5 on September 22, 2026, marking a strategic shift to deliver Fable-tier performance at lower costs.
Pricing for output tokens dropped to $20 per million and input tokens to $4 per million, representing reductions from the previous Opus 5 rates.
The update claims approximately 40% lower operational costs compared to its predecessor, alongside response speeds that are more than 30% faster.
On the GDPval-AA v2.1 benchmark, Opus 5.5 scored 1,846 points, outperforming both Opus 5 (1,708) and Fable 5.1 (1,735).
In Terminal-Bench 4.0 coding tests, the model achieved a score of 66.4%, surpassing OpenAI’s GPT-6 Astra which scored 57.9%.
Strict safety guardrails aligned with CEO Dario Amodei’s ‘pace the frontier’ philosophy are enforced, including permanently enabled extended thinking features.

A Strategic Shift in Pricing and Capability

Anthropic has officially released Claude Opus 5.5, marking a significant pivot in how its flagship AI model is priced and positioned within the enterprise market. Released on September 22, 2026 [1], this update delivers performance metrics that rival or exceed Anthropic’s own larger “Fable” tier models while simultaneously reducing operational costs by approximately 40% compared to its predecessor, Opus 5 [3].

This launch is not merely a technical iteration; it represents a strategic response to CEO Dario Amodei’s recent public call to “pace the frontier.” In this framework, Anthropic advocates for slowing the pace of raw capability advancement until safety and alignment protocols can catch up [1]. By offering Fable-level performance at Opus pricing, the company aims to provide high-end utility without accelerating unaligned capabilities unchecked [5].

The Cost-Performance Ratio

The most immediate impact of Opus 5.5 is financial for businesses relying on large language models (LLMs) for complex tasks. An LLM is a type of artificial intelligence trained to understand and generate human-like text, often used for coding, analysis, and customer service.

Anthropic has adjusted the pricing structure significantly:

Output tokens are now priced at $20 per million, down from $25 in Opus 5 [1]
Input tokens have dropped to $4 per million, a reduction from the previous $5 rate [3]

When combined with improved speed, Anthropic claims that running typical workloads on Opus 5.5 costs roughly 40% less than on Opus 5 [6]. The model also generates responses more than 30% faster due to optimized compute requirements [4]. This efficiency gap is critical for enterprise applications where latency and volume drive up costs.

Previously, the narrative around Anthropic’s enterprise readiness focused heavily on safety and reliability over raw speed or cost. However, Opus 5.5 directly addresses a common complaint about its predecessor: excessive jargon. The new model writes more clearly and places critical information at the beginning of responses, making it more usable for non-technical stakeholders [1].

Benchmark Performance vs. Peers

Opus 5.5 is positioned as the most capable tier in Anthropic’s lineup, sitting above Sonnet and Haiku models [4]. Its performance metrics suggest it can outpace competitors while maintaining a lower cost basis.

On the GDPval-AA v2.1 benchmark, which tests real-world work across 44 different occupations, Opus 5.5 scored 1,846 points [4]. This surpasses both its predecessor, Opus 5 (1,708), and Anthropic’s own Fable 5.1 model (1,735).

In coding-specific tests like Terminal-Bench 4.0, Opus 5.5 achieved a score of 66.4%, exceeding OpenAI’s GPT-6 Astra, which scored 57.9% [4]. However, GPT-6 Astra still led in specific science-focused subsets, indicating that no single model dominates every niche.

Anthropic cautions users that benchmark margins are narrowing at this high level of capability. The real-world performance gap between Opus 5.5 and Fable 5.1 is likely narrower than the scores suggest, meaning businesses may not need to pay a premium for the “Fable” tier if they only require top-tier reasoning [4].

Safety Guardrails and “Pacing the Frontier”

Shortly after CEO Dario Amodei urged the industry to slow down AI development to match safety progress [5], Opus 5.5 reflects this philosophy with strict safeguards similar to those applied to Anthropic’s restricted “Mythos” tier models.

Because Opus 5.5 possesses capabilities comparable to Fable and Mythos models in biology and cybersecurity, its use is limited regarding the discovery of exploits or the development of biological weapons [1]. The model achieved the highest scores on Anthropic’s automated behavioral audit to date, suggesting that increased capability has not compromised safety protocols [6].

Extended thinking features—a mechanism where the AI takes more time to process complex logic before responding—are permanently enabled and cannot be disabled by users [3] extended thinking features—a mechanism where the AI takes more time to process complex logic before responding—are permanently enabled and cannot be disabled by users [3]. This ensures consistent performance and safety monitoring across all interactions.

Availability and Future Roadmap

Opus 5.5 is available immediately via the Anthropic API on major cloud providers, including AWS, Google Cloud, and Microsoft Azure [3].

This launch serves as the first entry in the Claude 5.5 family. Sonnet 5.5 and Haiku 5.5 are scheduled for release in the coming weeks, promising similar improvements in speed, cost, and clarity across Anthropic’s entire product line [1].

For enterprises evaluating AI infrastructure, Opus 5.5 offers a compelling case study: higher performance at lower costs, wrapped in stricter safety constraints. It signals a mature phase for LLMs where efficiency and alignment are becoming just as important as raw intelligence.

Sources

Anthropic releases Opus 5.5 with lower prices and Fable-level performance | TechCrunch (techcrunch.com) — 2026-09-22
Claude Opus 5.5 vs Opus 5: Benchmarks and Cost (www.cosmicjs.com) — 2026-09-22
Anthropic launches Opus 5.5, its first model since CEO Amodei called for AI slowdown (finance.yahoo.com) — 2026-09-22
Introducing Claude Opus 5.5 (www.anthropic.com) — 2026-08-31
Anthropic launches Claude Opus 5.5: Benchmarks, pricing, safety (mashable.com) — 2026-09-22