Claude Fable 5.1 logo

Claude Fable 5.1

Anthropic · Sorti sept. 2026

Conditionnel

Anthropic's new frontier flagship is the most capable model publicly available, and it largely fixes the safeguard over-triggering that made Fable 5 frustrating for routine coding. The verdict stays conditional on cost: independent measurement shows more spend per task than Fable 5 despite the cheaper cache reads, because the model thinks longer. Reach for it on demanding reasoning and long-horizon agentic work. For everyday coding, Claude Opus 5 at half the rate remains the better default.

Est-ce fait pour vous ?

Recommandé pour

  • Demanding reasoning and long-horizon agentic work — Anthropic's own guidance is to escalate here when Opus 5 at higher effort still falls short
  • Agentic scientific research: 52.6% on Terminal-Bench-Science 0.1 against Opus 5's 29.0% and Fable 5's 24.7%
  • Context-heavy agent loops — cache reads dropped 75% to $0.25 per 1M, which Anthropic puts at up to ~45% cheaper on highly agentic workloads
  • Long-document and whole-repo work: 1M-token context with 128K max output
  • Security-adjacent coding that Fable 5 refused — cyber safeguards now permit vulnerability discovery, cutting Claude Code interventions ~60% per session

Déconseillé pour

  • Cost-sensitive everyday coding — Claude Opus 5 is 3 index points behind at $5/$25, half the rate
  • Latency-sensitive paths — adaptive thinking is always on and Anthropic rates its comparative latency as the slowest in the lineup
  • Budgeting by rate card alone — Artificial Analysis measures ~20% higher cost per task than Fable 5 despite the cache-read cut, because effort drives token count
  • Fact-recall work where a confident wrong answer is expensive — at max effort it attempts 93.4% of AA-Omniscience questions, trading more hallucinations for its higher accuracy at no net Omniscience gain

Performances par tâche

Agentic coding

Excellent

55.8% Terminal-Bench 4.0 and 73.4% CursorBench 3.2.0, with far fewer safeguard interruptions than Fable 5

Scientific research agents

Excellent

52.6% Terminal-Bench-Science 0.1, roughly double Fable 5 and well ahead of Opus 5's 29.0%

Frontier reasoning

Excellent

Tops the AA Intelligence Index at 66; 65.0% on Humanity's Last Exam with tools

Long-context work

Excellent

1M-token window, 128K output, and cache reads at $0.25 per 1M make repeated large-context passes affordable

Computer use

Good

41.7% OSWorld 2.0 strict — competent but not the reason to pay this rate

High-volume cheap inference

Poor

Wrong tier entirely; Haiku 4.5 at $1/$5 or Sonnet 5 at $2/$10 own this

Tarifs

Entrée

$10 / 1M tokens

Sortie

$50 / 1M tokens

Contexte

1M context, 128K output; cache $0.25/1M

Voir tous les tarifs

Tests de performance

TestScoreSource
Artificial Analysis Intelligence Index66 Source
Terminal-Bench-Science 0.152.6% Source
Terminal-Bench 4.055.8% Source
CursorBench 3.2.073.4% Source
Humanity's Last Exam (no tools)60.9% Source
Humanity's Last Exam (with tools)65.0% Source
AutomationBench31.4% Source
OSWorld 2.0 (strict)41.7% Source
GDPval-AA v21853 Source
AA-Omniscience accuracy (max effort)67.2% Source

Aucun changement de verdict pour l’instant

L’horloge tourne dès le premier jour : les changements apparaîtront ici à mesure que notre verdict évolue.

Journal de vérification

Aucune vérification pour le moment

Nous n’avons pas encore enregistré de vérification pour cette entrée. Dès qu’une vérification s’exécute, son historique apparaît ici.

Comment nous évaluons