Claude Logs Update Register

Current Snapshot

Checked on September 19, 2026: official Claude Code and Agent SDK docs continue to document persisted sessions, JSONL transcript storage under ~/.claude/projects/, hook lifecycle events, stream-json CLI output, OpenTelemetry metrics/logs/traces, identity attributes, and cost-tracking caveats. The transcript and telemetry surfaces this site describes have not changed shape.

What has changed is the model layer that transcripts record. The generally available lineup is Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5, and Claude Haiku 4.5, and every page, pricing row, and cost estimate on this site now reflects that.

This page is intentionally built as an extendable register. Future automation can add weekly entries without changing the core guide structure.

What To Watch

The most ranking-relevant changes for this domain are not model announcements. They are changes to how session evidence is produced: transcript persistence, background-session recovery, subagent transcript behavior, hook schema, detailed MCP logging, retry events, prompt/response logging switches, and CLI output formats.

General model news belongs on Claude Weekly. Deep benchmark interpretation belongs on Claude Reports. This register only tracks changes that affect logs and observability.

• Session and transcript storage behavior.
• Hook event schemas and lifecycle cadence.
• OpenTelemetry metric, event, trace, and identity fields.
• Stream-json and resume/naming CLI behavior.
• Cost and usage field semantics.

Entry - September 19, 2026

Refreshed the model layer across the site. The current generally available lineup is Claude Fable 5.1 (claude-fable-5-1), Claude Opus 5 (claude-opus-5), Claude Sonnet 5 (claude-sonnet-5), and Claude Haiku 4.5 (claude-haiku-4-5). Legacy Fable 5, Opus 4.x, and Sonnet 4.x identifiers still appear in older transcripts and remain in the token analyzer as separate pricing rows.

Three changes matter for cost review specifically. Cache reads on Fable 5.1 and Mythos 5.1 price at 2.5% of base input rather than the 10% that applies to the rest of the lineup, which can make a cache-heavy session far cheaper than its raw token count suggests. Sonnet 5 stayed at $2/$10 per million tokens instead of rising on September 1, so any estimate built against a $3/$15 Sonnet 5 row was wrong. And Fable 5.1 defaults to high effort in Claude Code but medium on Claude.ai and Cowork, so the same prompt can produce different token profiles by surface.

Two changes matter for transcript reading. Adaptive thinking is on across the current lineup, and the Fable 5.1 notes record that earlier models cannot read its thinking blocks and that editing an earlier turn invalidates thinking for that turn. A transcript replayed through a different model is therefore not the same artifact as the original.

Added containment evidence in response to Anthropic’s September 9 alignment assessment of four evaluation incidents. The page covers why a declared sandbox is not a control, why automated transcript review should score tool calls before reasoning text, and what a containment evidence set contains. It carries no attack detail.

Added the Black Box, a labeled synthetic Claude Code JSONL specimen played turn by turn with a redaction overlay and a containment readout. It is original work by Claude Logs, not a real private transcript. The existing log viewer remains the local-file tool.

• Updated the token analyzer pricing table, auto-detection, and pricing date to 2026-09-19.
• Added model-field and thinking-block guidance to transcript anatomy.
• Added model as a required dashboard dimension on the observability page.
• Added four primary sources: models overview, Fable 5.1 announcement, Opus 5 announcement, and the alignment assessment.
• Added the Black Box signature experience and a downloadable synthetic specimen.

Entry - July 6, 2026

Built the first Claude Logs sitemap and source index. Verified the domain lane against Cohort B rules: session-log analysis, transcripts as artifacts, and observability of agent runs. Added T4 treatment with AI crawler files, JSON-LD, citeable callouts, source blocks, and cite-this-page sections.

Next weekly review should diff the official Claude Code changelog and re-check docs pages for monitoring, hooks, CLI reference, session storage, and Agent SDK observability.