Claude Code 2.1.257 and 2.1.258: Fable 5.1, fixes

TL;DR: Claude Code 2.1.257 (released September 1, 2026) makes Claude Fable 5.1 the default Fable model, and 2.1.258 fixes the scheduled-session failure; gateway users must select 5.1 explicitly.

Two patches in one day, and they pull in opposite directions: 2.1.257 moves the default model forward while 2.1.258 repairs what the rollout broke. Anyone on 2.1.255 through 2.1.257 should read the second half first. This recap follows the official Claude Code changelog entry for entry, with the gateway trap most summaries skip.

What is new in 2.1.257 and 2.1.258?

1. Claude Fable 5.1 becomes the default Fable model. The new claude-fable-5-1 model ships with 1M context at $10/$50 per Mtok and takes over the default slot. Pricing context for what that means per session sits in the Claude Code pricing breakdown.
2. Time format and timeZone settings land. Sessions can now pin how timestamps render, which matters for scheduled sessions and log correlation across zones.
3. CLAUDE_CODE_SUBAGENT_MODEL_FORCE override appears. Operators can force the subagent model from the environment, useful for pinning cost or capability on delegated work.
4. Scheduled and remote sessions fixed in 2.1.258. Sessions that failed with user messages must have non-empty content after a re-sent permission approval now resume cleanly. If you disabled scheduled runs to dodge this, re-enable after upgrading.
5. macOS 12 (Monterey) launch regression fixed in 2.1.258. The regression introduced in 2.1.255 that stopped the CLI launching on Monterey is closed; 2.1.258 is the first safe upgrade for held-back Monterey machines.

Why does the Fable 5.1 default matter most?

The default-model switch is the only item that changes behavior without any action. From 2.1.257 on, a fresh session that resolves the Fable alias gets 5.1 with its 1M context window instead of Fable 5. Long-context work (repository-scale refactors, multi-file reviews) gains headroom, and per-token cost follows the $10/$50 schedule rather than the older blend. Nothing in existing prompts needs rewriting; the model behind the alias simply improves, which is exactly why the gateway exception below catches people out.

The trap: behind a Claude apps gateway, fable and best keep resolving to Fable 5 until the gateway itself is configured for 5.1. The changelog states this explicitly. So a gateway user who upgrades the CLI to 2.1.258, runs /model, and sees the old model is not experiencing a bug - the gateway alias table is stale. Pick Fable 5.1 explicitly in /model or update the gateway configuration, then verify with a version-reporting prompt before assuming the upgrade applied.

What breaks in this release?

Two regressions shipped inside the rollout window and both are repaired in 2.1.258. First, remote and scheduled sessions failed with user messages must have non-empty content whenever a permission approval was re-sent; the empty user message poisoned the turn and the run never recovered. Second, the 2.1.255 build would not launch at all on macOS 12 Monterey. Neither has a workaround worth preserving: the fix is the upgrade. The install paths per OS are covered in the Claude Code install guide, and the 60-second approval-timeout sibling fault is documented separately in the AskUserQuestion timeout fix. Both faults are filed in the Automation Error Index with other agent-tooling errors.

How do you upgrade?

Run the standard update path for your install and confirm the version string reads 2.1.258 before re-enabling anything you paused:

claude --version
# native install: follow the updater prompt, then re-run claude --version
# expect: 2.1.258 (or newer within the 2.1.x line)

Then verify in order: launch once on the affected machine (Monterey check), trigger one scheduled session manually (empty-content check), and run /model to confirm which Fable build the alias resolves to (gateway check). If a newer 2.1.x patch exists by the time you read this, take that instead - the guidance below holds for the whole line.

What is worth waiting for?

The open issue #63147 (62 comments, 50 reactions) reports that resuming an extended-thinking session fails permanently with 400: thinking blocks cannot be modified. It is separate from the two fixed regressions and has no shipped fix in the 2.1.257/2.1.258 pair. If your workflow leans on resuming extended-thinking sessions, track that thread before assuming 2.1.258 closes every resume-shaped fault.

FAQ

Should I upgrade to 2.1.258 immediately?

Yes, unless you are mid-incident on a pinned version. 2.1.258 contains no new behavior over 2.1.257 beyond the two regression fixes, so it is strictly safer than anything from 2.1.255 onward. Monterey machines held at older builds can move now.

Why does /model still show Fable 5 after I upgraded?

You are behind a gateway whose alias table still maps fable and best to Fable 5. Select claude-fable-5-1 explicitly in /model or reconfigure the gateway for 5.1; the CLI upgrade alone cannot change gateway-side resolution.

My scheduled sessions still fail with "user messages must have non-empty content" - did the fix miss?

Confirm the running version is 2.1.258 first; the fix is not in 2.1.257. If the version is right and the string persists, the failure is a different fault (for example approval-timeout behavior) rather than the repaired regression - capture the full turn log before reporting.

Claude Code will not launch on macOS 12 even after updating - what now?

Verify with claude --version that the binary actually updated; package-manager caches sometimes hold the old build. If 2.1.258 is confirmed and launch still fails, the cause is outside this regression and worth a fresh issue with the launch log.

Does the thinking-blocks resume error affect scheduled sessions too?

No evidence of that. Issue #63147 concerns resuming extended-thinking sessions interactively, a different path from the scheduled-session empty-content fault that 2.1.258 fixed. Track the issue separately.