Skip to content

Claude Fable 5

Released 2026-06 · thinking · 1.0M tokens · 5 benchmarks

Editorial notes

Lanzado 9 junio 2026. Primer modelo clase Mythos generalmente disponible (tier sobre Opus 4.8). Pricing $10/$50 per MTok, 1M context, 128K max output, adaptive thinking siempre activo (sin extended thinking). Safeguards de produccion: <5% de sesiones hacen fallback a Opus 4.8. Scores del System Card oficial Tabla 8.1.A (PDF leido directo, columna Fable 5 — adaptive thinking max effort, avg 5 trials): SWE-bench Verified 95, SWE-bench Pro 80, Terminal-Bench 2.1 84.3 (harness mini-SWE-agent; Mythos 5 sin safeguards: 95.5/80.3/88.0). HLE 53 medido INDEPENDIENTEMENTE por Artificial Analysis (Anthropic no publica HLE de Fable; solo Mythos 5: 59.0 sin tools). GPQA Diamond y AIME NO reportados para Fable — Anthropic los declara saturados y deja de publicarlos. #1 AA Intelligence Index (64.9, +5 sobre GPT-5.5). Otros: FrontierCode Diamond 29.3 (#1, eval Cognition), CursorBench 72.9 (medido por Cursor), GDPval-AA Elo 1932.

Spec sheet

Company
Anthropic
Country
US
Type
thinking
Release
2026-06
Context
1.0M tokens
Pricing (anthropic)
$10/$50/M
Slug
claude-fable-5

Benchmarks (5)

Cite this model
BibTeX · APA

BibTeX

@misc{frontier-claude-fable-5,
  title  = {Claude Fable 5},
  author = {{Anthropic}},
  year   = {2026},
  note   = {Frontier Benchmarks AI atlas. Accessed 2026-06-10},
  url    = {https://frontierbenchmarks.com/models/claude-fable-5}
}

APA

Anthropic (2026). Claude Fable 5 [Large language model]. Frontier Benchmarks AI. Retrieved 2026-06-10, from https://frontierbenchmarks.com/models/claude-fable-5

Citation reflects the atlas page, not the original model paper. For the paper, see the "Resources" section above.