TL;DR
Anthropic released Claude Opus 5.5 on September 22, 2026, a faster and cheaper flagship that beats its own larger Fable 5.1 model on key coding benchmarks, arriving just two months after Opus 5 and amid the company's own public calls to slow AI development.
What happened
- Claude Opus 5.5 launched September 22, 2026, two months after Opus 5 (July 24) and three weeks after the larger Fable 5.1 and Mythos 5.1 models.
- Pricing cut 20 percent on base rates: $4 per million input tokens and $20 per million output tokens, down from $5 and $25; cache reads slashed 60 percent to $0.20 per million tokens.
- Output generation is more than 30 percent faster than Opus 5, and Anthropic says the model costs roughly 40 percent less to run overall.
- Benchmark headline: 66.4 percent on Terminal-Bench 4.0 (vs. 55.8 percent for Fable 5.1 and 57.9 percent for GPT-6 Astra); 1,846 Elo on knowledge-work test GDPval-AA v2.1 (vs. 1,735 for Fable 5.1).
- Third-party safety audits by METR and Frontier Design cleared the model before release; Anthropic says it is 85 percent less likely than Opus 5 to attempt bypassing containment boundaries.
Why it matters
- A smaller, cheaper model outperforming a larger one on coding and knowledge work signals a new efficiency curve, compressing the cost-performance tradeoff that enterprises use to justify AI spend.
- The 60 percent cache-read price cut directly targets heavy coding and agentic workflows, where repeated context retrieval dominates cost, making Claude Opus 5.5 a direct threat to OpenAI's GPT-6 Sol (priced at $2 input / $10 output but trailing on Terminal-Bench 4.0).
- Independent verification is softer: Artificial Analysis scored Opus 5.5 at 59.6 percent on Terminal-Bench 4.0, not 66.4 percent, placing it level with GPT-6 Astra rather than ahead, and at max effort the model uses roughly 4x the output tokens of Astra.
- Anthropic's IPO timing: the release coincides with a widely anticipated fall IPO expected to be among the largest in history, making benchmark leadership commercially critical.
- The "pacing" contradiction is real: CEO Dario Amodei publicly called for slowing frontier AI development, yet Anthropic shipped two flagship-tier model families in under three months, a tension that regulators and investors will scrutinize.
What to watch next
- Sonnet 5.5 and Haiku 5.5 are promised "in the coming weeks"; their pricing and performance will determine whether Anthropic can undercut the mid-tier market where OpenAI's GPT-6 Luna competes.
- Independent benchmark replication: the gap between Anthropic's 66.4 percent and Artificial Analysis's 59.6 percent on Terminal-Bench 4.0 needs resolution before enterprises commit at scale.
- Anthropic's IPO prospectus: the company has signaled the document follows this launch, and its revenue figures will reveal whether the cost cuts are sustainable or a market-share grab ahead of going public.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.