Anthropic Launches Claude Opus 5.5 With Lower Prices and Better Agentic Coding Performance

Anthropic released Claude Opus 5.5 on September 22, 2026, positioning it as a flagship update focused on agentic coding, long-running knowledge work, and enterprise deployments. The company detailed the release in a post on its official site, and the launch was also covered by TechCrunch, which noted the model arrives at a lower price than its predecessor while matching or beating it on several fronts.

Lower Prices, Faster Output

The headline change for developers is cost. Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens, down from $5 and $25 respectively for Opus 5. Cache reads fall from $0.50 to $0.20 per million tokens, and cache writes drop from $6.25 to $5. Anthropic says typical workloads cost roughly 40% less, while output generation is more than 30% faster than Opus 5.

Benchmark Gains

Anthropic published several benchmark comparisons against Opus 5:

• Terminal-Bench 4.0: 66.4% for Opus 5.5 versus 52.3% for Opus 5

• FrontierCode v1.1: 54.4% versus 48.0%

• GDPval-AA v2.1: 1846 Elo versus 1708 Elo

• OSWorld 2.0 (partial credit): 81.8% versus 74.0%

Those benchmarks span coding-agent tasks, general knowledge-work evaluation, and computer-use scenarios, and the gains suggest Anthropic is optimizing this release specifically for agents that operate over longer horizons rather than single-turn chat.

Safety Testing and Prompt Injection Resistance

Anthropic said Opus 5.5 was evaluated by external groups including METR and Frontier Design ahead of release. The company described it as its “strongest-performing model we've tested to date” on an automated behavioral audit measuring how often a model takes hard-to-reverse or out-of-bounds actions.

Anthropic also reported that Opus 5.5 ties the larger Fable 5.1 for the lowest success rate against prompt-injection attacks in internal testing and carries the same cybersecurity and biology-related safeguards.

The model also includes an anti-distillation feature Anthropic calls “preserved thinking,” intended to make it harder for third parties to extract the model's reasoning traces for training competing systems.

Availability

Opus 5.5 is available now via the Claude API under the identifier claude-opus-5-5, as well as through Amazon Web Services, Google Cloud, and Microsoft Azure. Anthropic has signaled that updated Sonnet 5.5 and Haiku 5.5 models will follow within weeks, rounding out a refreshed 5.5 model family.

Context: A Crowded Release Week

The Opus 5.5 launch landed in the same 24-hour window as OpenAI's release of GPT-6 Sol and Luna, highlighting the increasingly rapid release cycle as both labs compete for developers building coding agents and high-volume automation tools.