Claude Opus 5.5 and Sonnet 5.5 Are Live on Scholarly
Claude Opus 5.5 and Claude Sonnet 5.5 are now in Scholarly's creation tools, and Auto (Best) now runs on Sonnet 5.5.
Anthropic released two new Claude models in the past week, and both are now on Scholarly:
- Claude Opus 5.5 (released September 22) replaces Claude Opus 5 as our Anthropic Frontier model.
- Claude Sonnet 5.5 (released September 28) replaces Claude Sonnet 5 in the Recommended Models group.
Both are creation models. You pick them when you turn your sources into something, such as a research report, a slide deck, a video lecture, a podcast, or an infographic. Neither one appears in AI Chat.
We also changed what Auto (Best) runs on. It now uses Claude Sonnet 5.5 instead of Claude Opus 5.5, at the same 3 AI Creation Credits. The rest of this post explains why.
Claude Opus 5.5: the Anthropic Frontier Model
Opus 5.5 takes the slot Opus 5 held in the Frontier Models group of the creation model menu, next to GPT 6 Astra. It costs 3 AI Creation Credits per creation, the same as GPT 6 Astra, and its row carries a 3× badge so you can see that before you choose it. Anthropic's list price for Opus 5.5 is $4 per million input tokens and $20 per million output tokens.
You can use it in:
- Deep Research
- AI Slides
- AI Video Lectures
- Podcasts
- Infographics
- Customization profiles, if you want a saved profile to always use it
If you had Claude Opus 5 selected anywhere, you don't need to do anything. Saved Opus 5 selections now use Opus 5.5 automatically.
Claude Sonnet 5.5: the New Recommended Claude Model
Sonnet 5.5 replaces Sonnet 5 under Recommended Models in the creation tools. It costs 3 AI Creation Credits per creation, up from 2 for Sonnet 5, and its row carries a 3× badge. Saved Sonnet 5 selections now use Sonnet 5.5 automatically. So do older Sonnet 4.6 selections, which were already being routed to Sonnet 5.
Like Sonnet 5 since September 13, Sonnet 5.5 is a creation-only model. If you want a Claude model in AI Chat, choose Claude Haiku 4.5, which is the Claude option there.
Why Auto (Best) Now Uses Sonnet 5.5
Auto (Best) is the option to choose when you want our strongest default for a creation and don't want to compare models yourself. It appears as Auto (Best) in the model menu and as the Use best model toggle in creation settings for research, slides, video lectures, podcasts, and infographics.
Until now it ran on Claude Opus 5.5 at low reasoning. Before switching it to Sonnet 5.5, we ran a blinded side-by-side test to see whether the cheaper model would give up quality.
The test. Both models wrote podcast scripts from the same four real professional sources:
- the NIST Cybersecurity Framework 2.0
- the AWS Cloud Practitioner exam guide
- Apple's FY2023 10-K
- a Japanese-language article
An independent judge scored every script on a 5-point rubric covering fidelity to the source, coverage, teaching quality, and presentation. The judge was not told which model wrote which script.
The result. Sonnet 5.5 at medium reasoning scored 3.00 and Opus 5.5 at low reasoning scored 2.94. A gap that small is effectively a tie. Sonnet 5.5 also cost roughly half as much to generate and was somewhat faster.
Since the quality was the same, we moved Auto (Best) to Claude Sonnet 5.5 at medium reasoning. Auto (Best) still costs 3 AI Creation Credits.
This is one test on one type of creation, and we will keep checking. It was, however, built on the kinds of material our customers actually use: a security framework, a certification exam guide, an annual report, and a source that isn't in English.
Credits
Paid plans come with a weekly allowance of AI Creation Credits that resets every Monday at 00:00 UTC: 10 on Premium and 80 on Laureate. Auto (Best), Claude Sonnet 5.5, Claude Opus 5.5, and GPT 6 Astra each cost 3 credits per creation. The plain Auto options still cost 1. A creation that fails returns its credits automatically. See Choosing an AI Model for the full list of models and their costs.
Opus 5.5 Is Still There If You Want It
We didn't remove Opus 5.5. We only stopped making it the default behind Auto (Best). If you prefer how Opus writes, or you want a second version of a report or deck from a different model, choose Claude Opus 5.5 by name from Frontier Models. It costs 3 credits, as before.
A Note for Teams
Owners and Admins on Scholarly Teams control which models their team can use. If your team has turned on Require Zero Data Retention, members will not see Claude Sonnet 5.5 or Claude Opus 5.5. Both are classified as provider-retention models, so the rule blocks them. The rest of the zero-retention catalogue is still available, and in the model catalogue each blocked model shows the rule that blocked it. See Team data and security for details.
Teams that don't require Zero Data Retention get both models the same way individual paid plans do.
How to Try Them
- Open Scholarly and start a research report, slide deck, video lecture, podcast, or infographic from your sources.
- Turn on Use best model to get Auto (Best) on Sonnet 5.5, or open the model menu.
- Choose Claude Sonnet 5.5 under Recommended Models, or Claude Opus 5.5 under Frontier Models.
Older blog posts and release notes still refer to Opus 5 and Sonnet 5 because they describe what was available when they were written.
