The AI model market is moving fast this September. On September 22, 2026, Anthropic released Claude Opus 5.5, the first model in the new 5.5 family. The same day OpenAI shipped the cheaper GPT-6 Sol and GPT-6 Luna, and earlier in the month came Claude Fable 5.1, GPT-6 Astra and Gemini 3.8 Flash. What matters for businesses is that top-tier performance keeps getting cheaper.
What Opus 5.5 brings
According to available information, Opus 5.5 delivers Claude Fable 5.1-level performance on most tasks while costing roughly 40% less to run than the previous Opus 5. Pricing is $4 per million input and $20 per million output tokens, 20% lower than before. Cache reads are 60% cheaper ($0.20 per million tokens) and output is more than 30% faster. A fast mode at $8 and $40 runs up to 2.5 times quicker. The model has a 1 million token default context and can produce up to 128,000 output tokens.
On Terminal-Bench 4.0, which measures command-line work and autonomous programming, it scored 66.4%. For comparison, GPT-6 Astra scores 57.9% and Claude Fable 5.1 scores 55.8%. Benchmarks should always be taken with a grain of salt, since they are not your specific project, but the direction is clear.
Why website owners should care
Cheaper and faster models show up in everyday things. A website chatbot answering customer questions can cost less to run. Automatic product descriptions, translations, transcripts or inquiry triage stop being an expensive toy. And if you run your own app or shop where AI does some of the work, your monthly bill drops.
The large context (1 million tokens) means the model can read big codebases or documentation in one go. For developers that means better handling of large projects, for example when refactoring a plugin or migrating a theme.
What it means for developers
AI coding assistants are now a routine part of the job. Faster and cheaper models make it practical to use them more often: for code review, test generation, bug hunting and documentation. The rule still stands that a human must check the output. A model can propose a solution, but responsibility for security, performance and accessibility lies with the team delivering the site. Unreviewed generated code is one of the most common ways vulnerabilities get into a website.
What to watch out for
- Data protection. Do not paste passwords, credentials or customer personal data into AI tools without checking the processing terms.
- Cost predictability. Set usage limits so automation does not grow beyond budget.
- Vendor independence. The market changes every few weeks, so build integrations so the model can be swapped.
Conclusion
Opus 5.5 is another step toward good AI being affordable for smaller companies too. The point is not to chase every new model but to pick the right tasks where it saves you time and money. If you are thinking about an AI chatbot, inquiry automation or speeding up work on your website, get in touch. At Full-Net we will help you choose a sensible solution and deploy it safely.
