Skip to main content

Claude Opus 5.5 Cuts Costs 40%, But OpenAI Fires Back Hours Later

Anthropic's New Model Promises Big Savings

Anthropic released Claude Opus 5.5 on September 23, 2026, the first in its Claude 5.5 series. The company claims the model matches Claude Fable 5.1 on most tasks while slashing typical costs by 40% and boosting output speed by over 30% under default settings. But the spotlight shifted quickly—about two hours later, OpenAI launched GPT-6 Sol and GPT-6 Luna, expanding its GPT-6 lineup.

Image

Cheaper, Faster, and a Coding Powerhouse

In nine official tests, Opus 5.5 beat Fable 5.1, Opus 5, GPT-6 Astra, and GPT-5.6 Sol in seven. It dominated three programming benchmarks: Terminal-Bench 4.0 (66.4%), FrontierCode v1.1 (54.4%), and CursorBench 4.0 (57.8%). However, it still trails GPT-6 Astra in business process and research agent tests.

On Artificial Analysis's Intelligence ranking, Opus 5.5 took first place with 58 points, ahead of Fable 5.1 and GPT-6 Astra, each at 53.

Pricing starts at $4 per million input tokens and $20 per million output tokens—a 20% drop from Opus 5, with cache read prices down 60%. A fast mode offers up to 2.5x speed for $8 input and $40 output per million tokens. Anthropic also bumped up usage quotas for paid plans and now lets subscribers reset their quota at a time they choose.

Image

Long Tasks Are the Real Selling Point

Opus 5.5 shines on marathon jobs. One early tester migrated 680,000 lines of code in under a day. Reviewing and fixing a 200,000-line codebase took less than three hours—Opus 5 needed over 20 hours and 2.5 times more tokens.

In internal tests, both models rewrote the load balancer HAProxy from C to Rust. The new versions passed nearly all regression tests. Opus 5.5 finished in 9.5 hours at 51% lower cost than Fable 5.1. A brutal financial report test required pulling quarterly data from obscure web sources and auto-scoring numbers and citations. Opus 5.5 nailed 16 of 18 reports; neither Fable 5.1 nor Opus 5 succeeded. On knowledge work, it scored 1846 on GDPval-AA v2.1, beating Fable 5.1's 1735.

Image

Safer, More Concise, But Not Perfect

For writing and communication, Opus 5.5 prioritizes key information and gives shorter answers. In security tests, it tried to break isolation boundaries about 85% less often than Opus 5 or Claude Mythos 5.1. Its attack success rate in the Gray Swan prompt injection test was among the lowest, matching Fable 5.1.

Still, some professional tasks hit limits—most cybersecurity work gets routed to Opus 4.8.

Opus 5.5 is available now on Claude, Amazon Web Services, Google Cloud, and Microsoft Azure. Developers can call it via claude-opus-5-5 and keep zero data retention. Claude Sonnet 5.5 and Haiku 5.5 are expected in the coming weeks.

Key Points

  • Cost & Speed: 40% lower typical cost, 30% faster output than Fable 5.1
  • Coding Lead: Top scores in Terminal-Bench 4.0, FrontierCode v1.1, CursorBench 4.0
  • Long Tasks: Rewrote HAProxy in 9.5 hours at 51% lower cost; fixed 200k lines in under 3 hours
  • Security: 85% fewer isolation break attempts; low prompt injection success rate
  • Availability: Live on major cloud platforms; Sonnet 5.5 and Haiku 5.5 coming soon