Anthropic sort Claude Opus 5.5, 40 % moins cher qu'Opus 5
TL;DR
- Grok 4.7's $2/$6 rate is the specific competitive trigger for the price move, per Forkast's sourcing, making this a market response rather than a pure efficiency dividend.
- Opus 5.5 is the first Opus model to ship with Fable-class bio and cyber safeguards, using transparent model fallback when gates trigger, a configuration dependency operators must audit before migration.
- Early testers completed 200,000-line codebase migrations in 3 hours; 9to5Mac cites a 680,000-line migration in under a day, confirming benchmark gains carry through to production workloads.
Anthropic met Claude Opus 5.5 en ligne le 22 septembre 2026, facturé 4 $ le million de tokens en entrée et 20 $ en sortie, soit 40 % de moins qu'Opus 5.
C'est le premier modèle d'une nouvelle famille 5.5. L'annonce d'Anthropic est directe: « It performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5. » Sur Terminal-Bench 4.0, Opus 5.5 marque 66,4 %, contre 52,3 % pour Opus 5 et 55,8 % pour Fable 5.1. La génération de sortie est annoncée 30 % plus rapide qu'Opus 5.
La partie sûreté occupe plus de place que d'ordinaire. Sur son audit comportemental automatisé, Anthropic décrit Opus 5.5 comme « the strongest-performing model we've tested to date ». Sur une nouvelle évaluation dédiée aux tentatives de franchir le périmètre imposé, le modèle « attempted to circumvent boundaries around 85% less often than Opus 5 or Claude Mythos 5.1 ». L'entreprise ajoute qu'il résiste mieux qu'Opus 5 aux injections de prompt.
Les accès sensibles passent par un gate. Les organisations validées peuvent postuler au Life Sciences Verification Program pour la recherche biologique. Côté cyber, l'entreprise applique « cybersecurity safeguards to Opus 5.5 that are similar to Fable 5.1's ». Frontier Design et METR ont conduit les évaluations externes avant lancement. Une protection anti-distillation baptisée « preserved thinking » est réservée aux comptes API créés après le 31 août 2026, une date qui prolonge la série d'actions d'Anthropic autour de la distillation que nous suivons cette semaine.
Ce qu'en disent les autres médias
-
AWS Blog Lire →
First-party deployment coverage: Bedrock regional availability across US, EU, AU, and JP CRIS inference profiles, Boto3 code examples, and CloudWatch/Cost Explorer integration for operators migrating now.
Claude Opus 5.5 does more with fewer tokens than Claude Opus 5, and new pricing passes those gains straight to customers.
-
9to5Mac Lire →
Carries a direct Anthropic quote linking the launch to its pacing-the-frontier commitment and the strongest-ever behavioral alignment audit score, plus a 680,000-line migration benchmark.
Opus 5.5 is our first model since we called for pacing the frontier. On our most comprehensive alignment test, it achieves the strongest score to date.
-
Forkast Lire →
Frames the price cut as a direct competitive counter to Grok 4.7's $2/$6 undercut and examines the tension between Anthropic's safety deceleration messaging and its aggressive market acceleration.
This pricing compression is a direct response to market pressures, specifically the $2/$6 undercut from Grok 4.7
-
MacRumors Lire →
Consumer framing: covers subscription tier limit increases and the October 22 rate-reset window alongside the API price move, surfacing the subscriber-side value for non-API Claude users.
Opus 5.5 is a 'major step up' from Opus 5 with improvements in performance and safety.
-
Techzine Global Lire →
European IT angle focused on ROI: flags that Opus 5 underperformed Fable in real-world use despite benchmark parity, and notes reduced 'Claudisms' as a measurable UX improvement.
An entire codebase can be migrated in a single day, with early testers completing a 200,000-line migration in three hours.
-
TestingCatalog Lire →
Developer-operator focus: specific token pricing ($4/$20 per million), 2.5x token efficiency gain per task, and a 200,000-line audit completing in 3 hours vs 20-plus hours with Opus 5.
Opus 5.5 performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.
Shared on Bluesky by 5 AI experts
-
-
Tim Kellogg @timkellogg.me
Opus 5.5 better and cheaper than Opus 5 — typically 40% cheaper, $4/mtok input, $20/mtok out www.anthropic.com/claude-opus-...
View on Bluesky →
Article original publié par anthropic.com
Lire l'article original →Titre original : Anthropic lance Claude Opus 5.5 : perf de Fable 5.1 pour 40 % de moins et safeguards renforcés