Claude Code release notes: 2.1.170 Fable 5 and saved transcripts

Claude Code 2.1.170 is not just another changelog line. It brings Claude Fable 5 into Claude Code, and it fixes a very practical problem: sessions launched from the VS Code terminal could fail to save transcripts and therefore not appear in --resume. For teams using Claude for code, documents, proposals, or internal analysis, both parts matter. A better model is useful. Traceable work is what makes the model safe enough to use in real workflows.

Claude Code release notes 2.1.170: new model plus saved transcripts

Claude Code is Anthropic's coding agent: an AI assistant that can work in a codebase, read files, propose edits, run commands, and continue work across steps in the terminal, IDE, web, or desktop. In release 2.1.170, Anthropic says Fable 5 becomes available by updating Claude Code to this version. The same release fixes sessions not saving transcripts, and not appearing in --resume, when launched from the VS Code integrated terminal or any shell that inherited Claude Code environment variables.

Source: Claude Code changelog

Source: GitHub release v2.1.170

That makes this signal bigger than a model announcement. If an agent does work for you, you need to see what it did, why it paused, and which conversation should be resumed. The transcript is not paperwork. It is the work receipt.

Fable 5 adds capacity, with visible fallback

Anthropic describes Claude Fable 5 as a Mythos-class model for general use. It uses the same underlying model as Claude Mythos 5, but Fable 5 has more conservative safeguards. For sensitive topics, a session can fall back to Claude Opus 4.8, and the user should be told when that happens. Anthropic also says Fable 5 is available in the Claude API as claude-fable-5, and that Pro, Max, Team, and seat-based Enterprise plans get temporary access from June 9-22, 2026 before continued use requires credits.

Source: Anthropic: Claude Fable 5 and Claude Mythos 5

For Nordic teams, the point is not to move everything to the newest model overnight. The point is to pick one task where stronger reasoning should be visible: a larger code change, a contract review, a requirements analysis, or an internal knowledge question where Claude has to keep many details in view. Log which model was used, whether fallback appeared, and what a human approved before anything moved forward.

What to check before you trust the result

Start with a small but real task. Save three things next to the result: the Claude Code version, the selected model, and the transcript or run log link. If the workflow touches internal systems, use scoped API keys, a secret manager, redacted logs, and a clear approval gate before the agent can write to production data. That is how AI becomes safely integrated instead of staying stuck in copy-paste mode.

There is also a small desktop detail: Anthropic had a resolved incident where Fable did not appear in the model picker in Claude Code Desktop. The status page says users who still see the issue should update to the latest version.

Source: Claude Status: Fable not available in the model picker in Claude Code Desktop

Try this prompt this week

Human step: Update or check the Claude Code version using your team's normal routine. Put the Fable 5 announcement, Claude Code 2.1.170 release, and the latest run log next to the task.

Read our Claude Code configuration, latest run log for [workflow],
and the links about Fable 5 and 2.1.170.
Write a short model test plan: task, selected model,
when fallback should be noted, which transcript should be saved,
and what needs human approval.
Suggest one first test. Do not change anything until I approve.

Good output should show:

• which model and version are actually being used
• where the transcript or run log is saved
• when fallback to another model is acceptable
• what a human must review before the next step

For Hammer Automation, this belongs in Tool Forge: connecting Claude to real tools, files, and systems with permissions, logs, and approvals that still make sense later.