Archived version · model page

Claude Fable 5.2

Version 3 of 3 · published 20 Nov 2026 · SHA-256 25f752c1

This is the latest published version. The live model page shows the same record. This page is frozen. It will not change when the live model page does, which is what makes it safe to cite.

What changed in this version

  • Benchmark scores: 1 updated

Record as published

Anthropic's 9 Nov 2026 Fable. Text and image in, text out. 1M context, 128K output. Knowledge cutoff Jun 2026. API list $8 / $40 per 1M tokens. Model ID claude-fable-5-2.

Provider
Anthropic
Family
Claude Fable
Role
Anthropic's model for demanding reasoning and long-horizon agent work. The docs say to start with Opus 5.5 and move here when that still falls short.
Status
flagship
Released
9 Nov 2026
Knowledge cutoff
1 Jun 2026
Context window (tokens)
1000000
Max output (tokens)
128000
Modalities
text, image, code, tool-use, reasoning
Access
api, consumer-app, managed-cloud
Input price (USD / 1M tokens)
8
Output price (USD / 1M tokens)
40
Pricing note
Claude API list checked 10 Nov 2026: $8 input / $40 output per 1M tokens. Cache reads $0.20. Five-minute cache writes $10; one-hour cache writes $16. Batch is half of standard. Fast mode is $16 / $80. US-only inference is 1.1×. The 1M window is standard price. Requests need 30-day retention unless Anthropic authorizes zero data retention. Default effort is high. Adaptive thinking stays on.
Pricing checked
10 Nov 2026
Editor's note
Recorded from the 9 Nov 2026 launch post, the Claude API model page, and the pricing page. Fast mode lists at $16 / $80. Requests need 30-day retention unless Anthropic authorizes zero data retention. Vendor benchmark tables stay off the ledger. The launch post points to a system card that was not on Anthropic's public system-card index on 10 Nov. Editorial depth still needs a desk pass.

Benchmark scores (1)

BenchmarkScoreSourceDate
aa-output-speed88.9 tpsArtificial Analysis20 Nov 2026

Sources

Checksum: SHA-256 over the previous version's hash and this record's canonical JSON. Any edit to a published version breaks the chain. Previous: 09de4056

Permanent URL

Cite this model page

Every format below points at version 3, published 20 Nov 2026. That URL shows exactly this record for as long as VerdictPal exists, so your reference needs no access date.

VerdictPal Editorial Desk. (2026, November 20). Claude Fable 5.2: VerdictPal model page (Version 3). VerdictPal. https://verdictpal.com/models/claude-fable-5-2/v/3

APA 7, web page with a group author. Paste into your reference list.

Version history

Each published version keeps its own URL and checksum. A later edit adds a new version; it never rewrites an old one.

  1. Version 3LatestYou are hereSHA-256 25f752c1
    • Benchmark scores: 1 updated
  2. Version 2SHA-256 09de4056
    • Benchmark scores: 1 added
  3. Version 1SHA-256 8ad229f2
    • First archived version.