AIが期間内の動向を整理
Anthropic resumes Fable 5, launches Sonnet 5, and unveils Claude Science
Anthropic resumed access to the paused Fable 5 and Mythos 5, and also announced Claude Sonnet 5 and the research-focused Claude Science. With Claude Tag for teams and follow-up details on safety standards, Claude’s use cases are expanding across everyday work, research, and operations.
SOURCE CHECK
Primary Sources 9
Primary Sources
Key Points
- 1Access to Fable 5 and Mythos 5 resumed on July 1
- 2Claude Sonnet 5 is strengthened for coding and professional work
- 3Claude Science is a workbench that brings research tools together
- 4Claude Tag helps teams organize conversations and work
- 5Safety measures and jailbreak evaluation guidance were also shared
Resuming paused models matters for real work
The return of Fable 5 and Mythos 5 is more than a simple feature update. When a model is unavailable, testing and comparison workflows tend to stall, but access resumed on July 1 across multiple surfaces. Some plans also have temporary benefits through July 7, making it easier to re-evaluate the models.
Sonnet 5 is easier to hand off longer and repetitive work
Claude Sonnet 5 is presented as stronger for coding, agents, and professional work. It is not just for short Q&A, but for repetitive development support and drafting operational work. In daily business use, it makes sense to start with smaller tasks and check the balance between quality and speed.
Claude Science supports research setup and reproducibility
Claude Science is a research workspace that brings together commonly used tools and packages, auditable outputs, and compute access. In research settings, setup and documentation are often bottlenecks, so reducing that overhead matters. It also makes it easier to preserve experiment logs and analysis steps, which helps reproducibility checks.
Team use and safety operations are being reviewed in parallel
Claude Tag is intended to organize conversations and workflows for groups so missed requests and handoffs are reduced. Anthropic also shared follow-up details on Fable 5 safeguards and the jailbreak scoring framework, making risk boundaries easier to understand. For adoption decisions, it is important to review not only performance but also who can use what and how it will be monitored.
Related News Items
Anthropic / feature
Delegate more everyday coding work to Claude
Anthropic / feature
Make scientific work easier to reproduce from notes
Anthropic / feature
Claude Science helps organize experiments and notes
Anthropic / feature
Claude Tag helps teams keep Claude work more organized
Anthropic / policy
Review harmful-use rules and safeguards earlier