TRACKING · Tracking record

Anthropic Claude · Records and changes

The Anthropic Claude model family, with distinct versions and evidence on pricing, capability, and safety.

00
UNDERSTANDING · What we know

Uses, conditions and important changes.

Each record has its own date and scope; these do not establish the live state of the whole product.

Claude Fable 5.1 · Reading revised · 5 Sept 2026

Claude Fable 5.1 tops the AA index; cache reads cut to 2.5% of input

Anthropic shipped Claude Fable 5.1, successor to Fable 5 for long-running agentic coding, knowledge work, and research, alongside Mythos 5.1 for Project Glasswing participants. Defaults: 1M-token context, 128k max output, always-on adaptive thinking at $10/$50 per MTok with prompt-cache reads cut to $0.25/MTok (0.025x base input versus 0.1x elsewhere). Live on Claude API, Bedrock, AWS, Google Cloud, Microsoft Foundry, and OpenRouter at matching prices. Artificial Analysis now ranks it #1 of 192 (index 66 vs 62 for Fable 5); cost per index task is $3.69 vs $3.14, about 18% higher.

Evidence is Anthropic's official release notes (vendor's own account) plus the Artificial Analysis leaderboard (third-party measurement, fetched and checked by BitShovel on Sep 2, 2026), cross-reported the same day by aihot and AlphaSignal. The ~18% per-task delta is computed from AA's $3.69 vs $3.14, slightly under the 20% rounding in aggregator headlines. An index rank is not a substitute for independent evaluation, and leaderboard attention is not proof of fit.

Read the record and sources →

Claude Opus 5 · Recorded by the site · 31 Aug 2026

Prompt injection plants a C2 callback inside Claude Code Auto Mode

wunderwuzzi (Embrace The Red) tested prompt injection against Claude Code Opus 5 Auto Mode: a hostile site baited the model into curl-in-Bash via a 415, then served a ZIP over a 303; the archive struct.py used module shadowing to execute when the model's decoder imported base64, establishing a Sliver C2 callback. Auto Mode allowed the malware process yet blocked Claude's cleanup. The author reports 60-80% success in 5-run samples, versus 0.00% in an Anthropic-commissioned eval; Anthropic closed the report as Informative: Auto Mode is a convenience feature, not a security guarantee.

Evidence is the author's own write-up (a creator self-report; body not in the feed snapshot, fetched and checked by BitShovel on 2026-08-31) plus the Lobsters front-page double reference. Not an independent review; 5-run success rates are not statistically conclusive, and front-page attention is not proof of generality.

Read the record and sources →

01
THE FULL RECORD · Full timeline

5 events retained by the site.

This object’s records stay together across live coverage, the watch pool and history. Every event links to its source record; corroboration and material changes have separate labels.

Dates use Beijing time. Open an event for its full record, original source and evidence. Corroboration adds support without counting as a material change.

02
COVERAGE · Record status

How these records are currently listed.

Coverage status describes how this site maintains its records, not whether a product is worth using. Existing records remain available after leaving the current list.

Registered follow-up scope

These are registered scopes, not complete product coverage or a receipt for the latest successful check.

  • Claude Fable 5.1

    Official revisions to Fable-series pricing, capabilities, or API behavior, plus any public widening of Mythos 5.1

    Open the tracked source ↗
  • Claude Opus 5

    Substantive revisions to the post, plus any Anthropic follow-up or Auto Mode security-behavior change

    Open the tracked source ↗

03
THE LINEAGE · Lineage

Every version on this line has its own full record.

The line's full timeline above is grouped by version; each version's own record opens from here.

View the line map