Claude Code v2.1.170 Major Update - Now featuring Mythos Class 1 Model Claude Fable 5!
This page has been translated by machine translation. View original
This is Ishikawa from the Cloud Business Division. Claude Code v2.1.170 was released on June 9, 2026. The biggest topic this time is support for the new model Claude Fable 5. Additionally, a fix has been included for a bug where session transcripts were not being saved.
Update Summary
v2.1.170 includes 2 changes. The highlight is that the new model Claude Fable 5 can now be used from Claude Code, and the other change is a fix for a bug where sessions were not being recorded when launched from the VS Code integrated terminal or similar environments.
Notable Updates
- New Feature: Support has been added for the new model Claude Fable 5 (Mythos class). It becomes available after updating to v2.1.170.
- Bug Fix: A bug has been fixed where transcripts of sessions launched from the VS Code integrated terminal and similar environments were not being saved and were not appearing in
--resume.
Update Details
New Feature: Support for Claude Fable 5
With v2.1.170, Claude Fable 5 can now be used from Claude Code. The CHANGELOG describes it as "a Mythos-class model made safe for general availability," and it becomes available by updating to v2.1.170.
Note that this change means Fable 5 has "become selectable," and does not mean the default model has been replaced with Fable 5.
Relationship between the Mythos Class, Fable 5, and Mythos 5
According to Anthropic's announcement, the relationship between the terms is as follows.
- Mythos class: A group of models positioned as a capability tier surpassing Opus. They were previously released as "Claude Mythos Preview" in April 2026.
- Claude Fable 5: A model that incorporates safety measures for general use into the Mythos class. This is the one that can now be used from Claude Code.
- Claude Mythos 5: A separately offered model with fewer restrictions. It is limited to cybersecurity partners (Project Glasswing) and a future program for biological researchers, and is not intended for use by general Claude Code users.
The following explanation focuses on Fable 5, which is available from Claude Code.
Key Capabilities
Anthropic describes Fable 5 as having "capabilities that surpass any model we've made generally available" and "state-of-the-art (SOTA) performance on virtually all benchmarks." The areas most relevant to Claude Code use cases are as follows.
-
Software Engineering: It is said to have high aptitude for complex migrations and refactoring. Stripe has reported that a migration of a Ruby codebase with 50 million lines of code, which was previously estimated to take about 2 months of manual work, was shortened to 1 day.
-
Vision (Image Processing): It is said to be state-of-the-art for tasks involving images, including reconstructing web apps from screenshots and reading numerical values from scientific charts.
-
Long Context: It is said to maintain focus in processing even at the scale of millions of tokens. This is a characteristic that is particularly effective for long agentic tasks and investigations of large codebases.
These are based on official announcements and partner reports, and actual effects will vary depending on the task. It is realistic to view this as potentially making it easier to conduct investigations spanning multiple files or long-running tasks with fewer interactions, as is typical with Claude Code.
CTO Luke Anderson commented that "Fable 5 achieves advanced engineering in fewer turns than previous models and handles the complex multi-agent workflows that employees run daily with Claude Code."
Claude Fable 5 is our first generally available Mythos-class model. New safety classifiers are built in, which may flag certain prompts in dual-use domains such as cyber and bio. Therefore, a fallback has been added so that rejected requests are retried with Claude Opus 4.8 instead of hitting a dead end.
Model ID and Pricing
- API identifier:
claude-fable-5 - Pricing: $10 per million input tokens, $50 per million output tokens
※ You will reach the limit approximately twice as fast as with Opus.
Since Claude Code frequently handles long contexts, rather than using it all the time, it may make sense to limit its use to high-difficulty migrations, refactoring, and similar tasks.
Availability Conditions and Safety Mechanisms (Notes for Use)
These are points to keep in mind before using it in practice. All are based on Anthropic's announcements.
- Availability conditions: Available via API and pay-as-you-go Enterprise. For subscription plans, it is included until June 22, 2026, and usage credits will be required from June 23, 2026 onward. Note that availability and billing methods vary depending on the contract type. (Take note if you are on the Claude Pro plan or Max plan.)
- Automatic routing of sensitive queries: Requests that fall under cybersecurity (offensive tasks), biology/chemistry, or model distillation attempts will be detected by classifiers and automatically routed to Claude Opus 4.8. This is said to be triggered in fewer than 5% of all sessions. It is worth understanding that the model responding may change for some requests.
- Data retention: For Mythos-class models, a 30-day data retention period is set for safety monitoring. If you handle highly confidential code or use it in an Enterprise context, we recommend verifying the relationship with existing usage policies and internal regulations.
Fix
- Fixed transcript not being saved in VS Code integrated terminal and similar environments: A bug has been fixed where session transcripts were not being saved and were not appearing in the
--resumelist when launched from the VS Code integrated terminal or a shell that inherited Claude Code's environment variables. This is a subtle but important fix for those who use--resumeto resume past sessions.
Closing
With v2.1.170, the new model Claude Fable 5 can now be used from Claude Code. While it is said to have high aptitude for complex migrations, refactoring, and long agentic tasks, it is also worth keeping in mind the practical prerequisites such as pricing, availability conditions, and automatic routing of sensitive queries. Additionally, the bug where sessions were not being recorded in the VS Code integrated terminal and similar environments has also been resolved. Why not update and give it a try?
Recent Updates
References