Claude Fable 5

Fiche modèle · Anthropic · claude-fable-5

Claude Fable 5 is Anthropic's generally available deployment of the same underlying weights as the restricted Claude Mythos 5 model. Anthropic positions the model as a new capability tier above Opus: Mythos 5 is the unsafeguarded trusted-access configuration, while Fable 5 adds production safeguards for public use. The system card reports that Mythos 5 is Anthropic's most capable model to date, with state-of-the-art results across software engineering, long-context agentic tasks, vision, knowledge work, healthcare, cyber, and life-sciences evaluations; Fable 5 matches the underlying model where safeguards do not trigger and behaves closer to Claude Opus 4.8 in safeguarded domains. Fable-specific results include SWE-Bench Verified 95.0%, SWE-Bench Pro 80.0%, Terminal-Bench 2.1 84.3% mean reward with 20.9% of trials hitting a safety fallback, OSWorld-Verified 85.0%, FrontierCode Diamond 29.3%, FrontierCode Main 46.3%, GDPval-AA 1932 Elo, GDP.pdf 29.8%, Blueprint-Bench 2 38.6%, and AutomationBench 17.4%. Fable 5 uses classifiers for cybersecurity, biology and chemistry, and distillation attempts; in Claude client apps these requests fall back to the latest Claude Opus model, while the Messages API blocks by default unless developers implement or opt into fallback. Separate competitive-use safeguards target frontier LLM development work and are estimated by Anthropic to affect about 0.03% of traffic. The model was trained on a proprietary mix of public web data, public and private datasets, and synthetic data. Anthropic prices Fable 5 at $10 per million input tokens and $50 per million output tokens, and exposes it through the Claude API as `claude-fable-5`.

Identité

OrganisationAnthropic
Pays—
Famille—
Typellm
Date de sortie2026-06-09
Licenceproprietary
Licence commercialenon
Licence open sourcenon
Poids ouvertsnon
Auto-hébergement (DGX Spark)non — poids fermés
Modalitésimage|text
Fenêtre de contexte1 000 000
KV cache / token (Q8)—
Paramètres totaux≈ 5 T (estimation, confiance très faible — https://www.reuters.com/technology/bytedance-targets-mega-ai-model-nearing-anthropics-mythos-ft-reports-2026-08-07)
Paramètres actifs (session)≈ 84 B–102 B (estimation, confiance très faible — https://www.reuters.com/technology/bytedance-targets-mega-ai-model-nearing-anthropics-mythos-ft-reports-2026-08-07)
Source des paramètres actifshttps://www.reuters.com/technology/bytedance-targets-mega-ai-model-nearing-anthropics-mythos-ft-reports-2026-08-07 (estimation, confiance très faible)
Curation des paramètres actifs2026-10-27
Prix par session20.00 $ (1 000 000 tokens entrée + 200 000 sortie, prix le moins cher par fournisseur)
Tokens / tâche (Data Players)1.2 M
Tokens / tâche (Terminal-Bench)57.6 M
Ratio Terminal-Bench / Data Players48.0×
Coût / tâche (Terminal-Bench)110.61 $
Tokens d'entraînement—
Coupure de connaissance—
Créé le2026-06-09T20:20:02.226462Z
Mis à jour le2026-11-04T06:04:32.380323Z
SourceLLM Stats
URLhttps://llm-stats.com/models/claude-fable-5

Prix par fournisseur

Statut du modèle (offres + auto-hébergement) : non.

Toutes sources confondues : catalogue llm-stats, routeurs EUrouter/Cortecs et catalogues directs (OVHcloud, Scaleway, Infomaniak). Les prix sont convertis en USD pour la comparaison ; le prix publié et le taux BCE utilisé sont dans l'infobulle. Les prix inconnus restent indiqués « — ».

FournisseurSourceSouverainetéEntrée (/M tok)Sortie (/M tok)QuantificationContexteLatence (s)Débit (tok/s)Statut
Anthropicllm-statsnon10.00 $50.00 $————active
DeepInfrallm-statsnon10.00 $50.00 $————active

Souveraineté qualifiée par offre (modèle × fournisseur) ; survolez le badge pour les critères.

Souveraineté

Scores par catégorie (top_scores)

CatégorieScore
legal0.10
math0.70
multimodal0.50
reasoning0.50
safety0.80
science0.50
finance0.60
frontend_development0.90
general0.60
healthcare0.70
agents0.60
biology0.50
code0.70
tool_calling0.50
vision0.60

Scores par benchmark

BenchmarkCatégorieScoreMaxNormalisé (API)VérifiéAuto-déclaréDaté du
automationbenchagents0.171.000.17nonoui2026-11-04T06:04:32.417007Z
biomysterybenchbiology0.461.000.46nonoui2026-11-04T06:04:32.447847Z
blueprint-bench-2agents0.391.000.39nonoui2026-11-04T06:04:32.413337Z
deepswe-1.1code0.701.000.70nonnon2026-11-04T06:04:32.483413Z
exploitbenchcode0.781.000.78nonoui2026-11-04T06:04:32.458692Z
finance-agent-v2agents0.561.000.56nonnon2026-11-04T06:04:32.470611Z
frontiercodereasoning0.461.000.46nonoui2026-11-04T06:04:32.403524Z
frontiercode-1.1reasoning0.541.000.54nonnon2026-11-04T06:04:32.493560Z
frontierswecode0.901.000.90nonnon2026-11-04T06:04:32.474126Z
gdp-pdfgeneral0.301.000.30nonoui2026-11-04T06:04:32.407921Z
healthbench-professionalhealthcare0.661.000.66nonoui2026-11-04T06:04:32.466965Z
humanity's-last-examreasoning0.651.000.65nonoui2026-11-04T06:04:32.440631Z
legal-agent-benchmarklegal0.131.000.13nonoui2026-11-04T06:04:32.428273Z
lifesci-diamond-2026-11healthcare0.131.000.13nonnon2026-11-04T06:04:32.503051Z
livebenchmath0.781.000.78nonnon2026-11-04T06:04:32.477604Z
osworld-verifiedmultimodal0.851.000.85nonoui2026-11-04T06:04:32.421069Z
swe-bench-proreasoning0.801.000.80nonoui2026-11-04T06:04:32.384731Z
swe-bench-verifiedreasoning0.951.000.95nonoui2026-11-04T06:04:32.393202Z
terminal-bench-2.1reasoning0.841.000.84nonoui2026-11-04T06:04:32.453143Z
terminal-bench-4.0reasoning0.451.000.45nonnon2026-11-04T06:04:32.499225Z

Classements

CatégorieRangScoreNote conservatriceBenchmarks évaluésMéthodeDaté duProvenance
agents120.6038.6813trueskill2026-11-04T08:08:47.839150Zrankings
code70.7043.769trueskill2026-11-04T08:08:39.742311Zrankings
math80.7042.022trueskill2026-11-04T08:08:42.442591Zrankings
multimodal280.5031.833trueskill2026-11-04T08:08:46.488887Zrankings
reasoning110.5050.3614trueskill2026-11-04T08:08:41.093494Zrankings

Lecture : Note conservatrice = borne basse TrueSkill (niveau estimé − marge d’incertitude) ; Benchmarks évalués = nombre de benchmarks de la catégorie pris en compte. À performances égales, un modèle évalué sur moins de benchmarks obtient une note plus basse : la note ne se déduit pas d’un seul score. Provenance : rankings = top 50 de /rankings ; model_history = historique du modèle (/models/{id}/history), utilisé pour les rangs au-delà du top 50.

Données

Dates de curation (récupération depuis l'API) : models 2026-11-04 · benchmarks 2026-11-04 · scores 2026-11-04 · updates 2026-11-04 · account 2026-11-04 · model_history 2026-11-04 · rankings 2026-11-04