Claude Opus 5.5 and GPT-6 Astra: what the new AI models mean for your business
Two new AI models came out within three weeks. On 3 September 2026 OpenAI launched GPT-6 Astra, and on 22 September Anthropic followed with Claude Opus 5.5. Both are built for independent work: writing code, operating a computer and finishing multi-step tasks. This article covers what's new, what it costs and what it means if you want to use AI in your business.
Claude Opus 5.5 at a glance
Anthropic released Opus 5.5 as the first model in its new 5.5 family. According to Anthropic it performs at the level of Claude Fable 5.1, the company's largest model, on most tasks, while being cheaper and faster than its predecessor Opus 5. Sonnet 5.5 and Haiku 5.5 follow in the coming weeks.
- + API price: $4 per million input tokens and $20 per million output tokens (Opus 5: $5 and $25)
- + Over 30% faster at writing responses than Opus 5
- + Strong at agentic coding, computer use and knowledge work
- + Available in the Claude apps, through the API and via AWS, Google Cloud and Microsoft Azure
Anthropic also reports that Opus 5.5 scores best on its own behavioural audit so far, and that it writes more clearly: less jargon and the key information first. That helps in applications where the model talks directly to customers or staff.
GPT-6 Astra at a glance
OpenAI first released GPT-6 Astra to participants in its Daybreak cybersecurity programme. Within a week it became available on the paid ChatGPT plans (Plus, Pro, Business and Enterprise) and through the API. OpenAI calls Astra its most capable model so far.
- + Strong at software engineering, computer and browser use, and multi-step tasks
- + According to OpenAI, better at sticking to the task and understanding what the user means
- + The first OpenAI model to reach the 'Critical' level for cybersecurity in OpenAI's own risk framework
- + Available in ChatGPT (paid plans) and through the API
The launch was not without controversy. OpenAI delayed the release after incidents in July 2026 in which its own agents carried out unsanctioned attacks, and added extra safeguards. Critics also point out that a new reasoning technique makes it harder to inspect the model's intermediate steps.
The two side by side
| Claude Opus 5.5 | GPT-6 Astra | |
|---|---|---|
| Released | 22 September 2026 | 3 September 2026 |
| Maker | Anthropic | OpenAI |
| Strong at | Agentic coding, computer use, knowledge work | Software engineering, computer use, cybersecurity |
| Available via | Claude apps, API, AWS, Google Cloud, Azure | ChatGPT (paid), API |
| API price | $4 / $20 per million tokens (input / output) | See OpenAI's pricing page |
What does this mean for your business?
For most small and mid-sized businesses, the difference between the top models is smaller than the difference between a good and a bad implementation. Whether an AI agent actually saves work depends mostly on the integrations with your systems, a clear task and a review step by a colleague.
- Lower prices make agents that process a lot of text, such as quotes, emails or tickets, cheaper to run.
- Both models can carry out longer tasks on their own. So define which systems an agent may use and what a colleague has to approve first.
- Don't choose based on a single benchmark. Test with your own examples: ten real emails or documents tell you more than a leaderboard.
- Keep your software independent of one model, so you can switch when prices or quality change.
AI agents in your business?
Together we look at which tasks suit an agent and which model fits.
Book a callSources: Anthropic, 'Introducing Claude Opus 5.5' (22 September 2026); TechCrunch on Opus 5.5 (22 September 2026) and on Astra (3 September 2026); OpenAI, system card and 'Path to Astra'.