On September 22, 2026, Anthropic released Claude Opus 5.5, the latest version of Claude's top-tier model. It is the first model in the Claude 5.5 family, and it stands out for improving performance while significantly lowering the cost of use.
Cost and speed
According to the official announcement, Opus 5.5 cuts per-token pricing by 20% for both input and output compared with Opus 5. Because it also completes the same work with fewer tokens, Anthropic says the overall cost of typical tasks drops by about 40% at default settings. Cache reads are also 60% cheaper.
Output generation is also more than 30% faster than with Opus 5.
Pricing
API pricing per million tokens is as follows.
- $4 for input and $20 for output
- $0.20 for cache reads and $5 for cache writes
- Fast mode: $8 for input and $40 for output
Key numbers
These are the main figures Anthropic published.
- 66.4% on Terminal-Bench 4.0, an agentic coding evaluation (Opus 5: 52.3%)
- 54.4% on FrontierCode v1.1, a coding evaluation (Opus 5: 48.0%)
- 1846 Elo on GDPval-AA v2.1, an evaluation of real-world work tasks (Opus 5: 1708)
- 81.8% on OSWorld 2.1, a computer-use evaluation (including partial credit; Opus 5: 74.0%)
All figures come from Anthropic's own announcement.
Other improvements
Anthropic says the model communicates more naturally and leads with the most important information, and has stronger safeguards against misuse in cybersecurity and biology. Its resistance to prompt injection was also equal to or better than Opus 5's in every condition tested.
Opus 5.5 is available on all platforms, including Claude Code. It can be used on the Claude Platform for developers (model ID: claude-opus-5-5), as well as on Amazon Web Services, Google Cloud and Microsoft Azure.
CoAI's take
Top-tier models deliver strong performance, but price has often been the sticking point. This price cut makes using Opus for everyday development and work a realistic option.
