{"id":1379,"date":"2026-06-07T18:50:13","date_gmt":"2026-06-07T13:20:13","guid":{"rendered":"https:\/\/stnews.live\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-2\/"},"modified":"2026-06-08T20:22:13","modified_gmt":"2026-06-08T14:52:13","slug":"claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6","status":"publish","type":"post","link":"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/","title":{"rendered":"Claude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faiblesses r\u00e9duites"},"content":{"rendered":"“`html\n\n

La critique de Claude Opus 4.8 : Meilleur dans ce qu’il fait bien, moins bon dans ce qu’il ne peut pas faire, montre que Claude Opus 4.8 arrive avec des benchmarks techniques r\u00e9els qui montrent des progr\u00e8s concrets dans les t\u00e2ches de code, de flux de travail et de prototypage, comme le rapporte Thezvi<\/a>. Ce saut dans la capacit\u00e9 de programmation r\u00e9elle \u2013 de 64,3 \u00e0 69,2 dans le benchmark Pro SWE-bench \u2013 refl\u00e8te combien de programmation de routine et de recherches mineures le mod\u00e8le g\u00e8re d\u00e9sormais, surtout par rapport aux LLM pr\u00e9c\u00e9dents, plus ax\u00e9s sur la correspondance de motifs. Et avec des prix fix\u00e9s \u00e0 5 $ pour les entr\u00e9es et 25 $ pour les sorties par million de tokens, Opus 4.8 permet aux utilisateurs d’obtenir plus de valeur quotidiennement sans co\u00fbts suppl\u00e9mentaires, selon Thezvi et Lennysnewsletter<\/a>.<\/p>\n\n\n

Cette critique de Claude Opus 4.8 traite de la mani\u00e8re dont le mod\u00e8le fournit des am\u00e9liorations significatives dans la programmation de routine, l’automatisation des flux de travail et le prototypage rapide. Cependant, des faiblesses persistantes dans les t\u00e2ches ambigu\u00ebs ou hautement strat\u00e9giques sont \u00e9galement d\u00e9crites en d\u00e9tail. Cela confirme le th\u00e8me selon lequel Claude Opus 4.8 est meilleur dans ce qu’il sait d\u00e9j\u00e0 bien faire, mais moins bon ou inchang\u00e9 l\u00e0 o\u00f9 il avait auparavant des difficult\u00e9s.<\/p>\n\n\n

\n\n\n

Principales perspectives de la communaut\u00e9<\/h2>\n\n\n

Analyse de Lesswrong et de la communaut\u00e9<\/h3>\n\n\n

La communaut\u00e9 Lesswrong<\/a> enregistre \u00e0 la fois des progr\u00e8s mesurables et une limite qualitative famili\u00e8re pour Claude Opus 4.8. Les participants aux classements publics et aux protocoles priv\u00e9s transmettent un message : Opus 4.8 m\u00e9rite vraiment ses scores Pro SWE-bench plus \u00e9lev\u00e9s. Mais les contributeurs notent souvent des “erreurs marginales” lorsque les r\u00e8gles sont floues ou que les instructions changent en cours de processus. Les transcriptions annot\u00e9es de LW mettent en \u00e9vidence les types d’ambigu\u00eft\u00e9 qui bloquent Opus 4.8 \u2013 logique conditionnelle non \u00e9vidente, raisonnements tortueux ou surprises s\u00e9mantiques.<\/p>\n\n\n

Les contr\u00f4les d’effort et les fonctions de sous-agents aident \u00e0 combler les lacunes dans les t\u00e2ches de base, mais ils ne peuvent tout simplement pas couvrir les capacit\u00e9s de planification n\u00e9cessaires pour une recherche rigoureuse ou une logique commerciale.<\/p>\n\n\n

Perspective commerciale de Lennysnewsletter<\/h3>\n\n\n

Lennysnewsletter se concentre sur les impacts r\u00e9els d’Opus 4.8 : il est configur\u00e9 pour le prototypage commercial rapide, le d\u00e9ploiement rapide de fonctionnalit\u00e9s et les flux de travail parall\u00e9lis\u00e9s sur Claude.ai et Cowork. Selon cette critique de Claude Opus 4.8, le lancement de prototypes Greenfield ou la r\u00e9alisation d’exp\u00e9riences ponctuelles est d\u00e9sormais plus facile.<\/p>\n\n\n

\n\n\n

Tester Claude Opus 4.8 : M\u00e9thodes et r\u00e9sultats<\/h2>\n\n\n

Cr\u00e9ation des tests<\/h3>\n\n\n

Les concepteurs de tests se sont appuy\u00e9s sur la validation crois\u00e9e pour les \u00e9valuations actuelles \u2013 en utilisant des protocoles et des \u00e9tudes de cas de Thezvi, Lennysnewsletter et Lesswrong pour garantir des normes tant pour les benchmarks publics que priv\u00e9s.<\/p>\n\n\n

R\u00e9sultats des tests<\/h3>\n\n\n

Le grand saut dans les scores Pro SWE-bench \u2013 de 64,3 \u00e0 69,2 \u2013 capture un v\u00e9ritable progr\u00e8s \u00e9tape par \u00e9tape, selon les protocoles de Thezvi. Cependant, les examinateurs s’accordent \u00e0 dire : la programmation de routine brille, mais d\u00e8s que l’ambigu\u00eft\u00e9 du test augmente, Opus 4.8 se bloque. L’ach\u00e8vement de code et l’automatisation r\u00e9compensent la mise \u00e0 niveau, tandis que les travaux lourds en strat\u00e9gie ou la planification approfondie continuent de r\u00e9v\u00e9ler des points de d\u00e9rive, selon les r\u00e9sultats de Lennysnewsletter. Le crowdsourcing de Lesswrong confirme une diminution des erreurs de syntaxe et des erreurs de type “off-by-one” dans des sc\u00e9narios pr\u00e9visibles, ce qui soutient les principales conclusions de cette critique de Claude Opus 4.8 : meilleur dans ce qu’il fait bien, moins bon dans ce qu’il ne peut pas faire.<\/p>\n\n

Learnings from testing Claude Opus 4.8:

> Much worse than Opus 4.7 and GPT 5.5 on Vending Bench
> More aligned than previous Claude models (Opus 4.6+ and Mythos)
> Also worse on Blueprint-Bench
> Scared of getting caught
> Max reasoning is not the best reasoning effort pic.twitter.com\/9yn58xsJL9<\/a><\/p>\u2014 Andon Labs (@andonlabs) May 28, 2026<\/a><\/blockquote>\n\n\n\n

\n\n\n

Benchmarks et comparaisons de Claude Opus 4.8<\/h2>\n\n\n

Chiffres de benchmark : 4.8 vs 4.7 vs GPT-5.5<\/h3>\n\n\n
\n\n\n\n\n\n\n
Mod\u00e8le<\/th>\nScore Pro SWE-bench<\/th>\nCo\u00fbts d’entr\u00e9e\/sortie par 1M de tokens<\/th>\nCo\u00fbts du mode rapide<\/th>\n<\/tr>\n<\/thead>\n
Claude Opus 4.8<\/td>\n69.2<\/td>\n5 $ \/ 25 $<\/td>\nMoins cher que 4.7 (30 $ \/ 150 $)<\/td>\n<\/tr>\n
Claude Opus 4.7<\/td>\n64.3<\/td>\n5 $ \/ 25 $<\/td>\n30 $ \/ 150 $<\/td>\n<\/tr>\n
GPT-5.5<\/td>\nNon disponible<\/td>\nNon divulgu\u00e9<\/td>\nNon divulgu\u00e9<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n

La documentation de Thezvi pr\u00e9cise qu’avec 5 $ pour les entr\u00e9es et 25 $ pour les sorties par million de tokens, Opus 4.8 atteint le prix de liste de 4.7, mais le sous-cote d\u00e9sormais en mode rapide \u2013 30 $ \/ 150 $ pour 4.7 \u2013 rendant ainsi les travaux par lots abordables et permettant une utilisation en direct pour les petites \u00e9quipes.<\/p>\n\n\n

\n\n\n

Qu’est-ce qui est r\u00e9ellement nouveau (au-del\u00e0 des chiffres)<\/h2>\n\n\n

Lennysnewsletter explique comment Claude.ai et Cowork sont d\u00e9sormais dot\u00e9s de mises \u00e0 jour beaucoup plus pratiques pour les d\u00e9veloppeurs. En particulier, le lancement de prototypes Greenfield ou la r\u00e9alisation d’exp\u00e9riences ponctuelles est d\u00e9sormais plus facile avec Claude Opus 4.8.<\/p>\n\n\n

5 $ \u2014 Co\u00fbt par million de tokens d’entr\u00e9e.<\/p>\n\n\n

\n\n\n

Claude Opus 4.8 vs. Sonnet 4.6<\/h2>\n\n\n

Devriez-vous utiliser Claude Opus 4.8 ou Sonnet 4.6 ?<\/h3>\n\n\n

Les tests de Lennysnewsletter montrent qu’Opus 4.8 est sup\u00e9rieur \u00e0 Sonnet 4.6 en mati\u00e8re de code de routine et de prototypage rapide. Pour une logique progressive ou de la clart\u00e9, beaucoup choisissent Sonnet, tandis que les efforts d’automatisation tendent d\u00e9sormais vers Opus. Pour la plupart des t\u00e2ches de flux de travail, cette critique de Claude Opus 4.8 conclut qu’Opus est le gagnant.<\/p>\n\n\n

\n\n\n

Claude Opus 4.8 vs. GPT-5.5<\/h2>\n\n\n

Comment Opus 4.8 se compare-t-il \u00e0 GPT-5.5 ?<\/h3>\n\n\n

Il n’y a pas de scores Pro SWE-bench publi\u00e9s pour GPT-5.5, ce qui cr\u00e9e une lacune pour la comparaison directe, comme le montre la documentation de Lesswrong. Et tandis que Thezvi et Lennysnewsletter clarifient les prix et les flux de travail d’Opus, les co\u00fbts et les gains de programmation r\u00e9els de GPT-5.5 restent dans une bo\u00eete noire. Les testeurs comparent ce qu’ils peuvent : les nouveaux flux de travail automatiques, la parall\u00e9lisation des sous-agents et la gestion des entr\u00e9es\/sorties d’Opus 4.8 \u2013 des fonctionnalit\u00e9s qui n’ont pas encore \u00e9t\u00e9 atteintes dans les benchmarks publics de GPT-5.5. Dans la planification commerciale intensive et la complexit\u00e9, les protocoles de Lennysnewsletter sugg\u00e8rent qu’aucun des deux ne surpasse constamment l’autre dans des strat\u00e9gies profondes. Opus 4.8 gagne en termes de vitesse, Sonnet en termes de clart\u00e9, GPT-5.5 en termes de largeur th\u00e9orique \u2013 mais jusqu’\u00e0 ce que les donn\u00e9es Pro SWE-bench arrivent, le classement reste ind\u00e9cis.<\/p>\n\n\n

Meilleurs cas d’utilisation pour Claude Opus 4.8<\/h3>\n\n\n

Toutes les trois principales sources \u2013 Thezvi, Lennysnewsletter et Lesswrong \u2013 pour Opus 4.8 s’accordent sur les m\u00eames applications : il est con\u00e7u pour accomplir des t\u00e2ches o\u00f9 votre flux de travail et vos objectifs restent clairs et mesurables. Comme le prouve cette critique de Claude Opus 4.8, il est le mieux adapt\u00e9 pour l’automatisation de routine, les t\u00e2ches de programmation, le d\u00e9veloppement rapide de prototypes et les flux de travail g\u00e9r\u00e9s.<\/p>\n\n\n

D\u00e9veloppements futurs et feuille de route d’Anthropic<\/h3>\n\n\n

Le prochain cycle d’Opus 4.8 \u2013 selon Thezvi \u2013 visera pr\u00e9cis\u00e9ment ces cas limites et probl\u00e8mes de contexte, ce qui sugg\u00e8re que les \u00e9valuations futures pourraient d\u00e9placer le r\u00e9cit dans “Critique de Claude Opus 4.8 : Meilleur dans ce qu’il fait bien, moins bon dans ce qu’il ne peut pas faire”.<\/p>\n\n\n

Principales conclusions<\/h3>\n\n\n

Les enregistrements de Thezvi confirment : le score de 69,2 de Pro SWE-bench d’Opus 4.8 consolide sa position de leader dans les t\u00e2ches de flux de travail structur\u00e9es et contr\u00f4l\u00e9es par le code. Le prix reste stable \u00e0 5 $ \/ 25 $ et un mode rapide moins cher est ajout\u00e9.<\/p>\n\n\n

R\u00e9action de la communaut\u00e9 et retours en direct<\/h3>\n\n\n

Les fils en direct de Lesswrong et les protocoles collaboratifs de Thezvi montrent des r\u00e9actions mitig\u00e9es : optimisme concernant l’automatisation, mais avertissements clairs que les limites du mod\u00e8le en mati\u00e8re d’argumentation n’ont pas chang\u00e9. Les chercheurs et les utilisateurs avanc\u00e9s s’accordent \u00e0 dire : des sauts de productivit\u00e9 dans la programmation et la routine automatis\u00e9e, mais des incertitudes dans les cas limites et les hallucinations suscitent la prudence. La programmation est plus coh\u00e9rente, mais, comme le r\u00e9p\u00e8tent plusieurs r\u00e9sum\u00e9s des critiques de Claude Opus 4.8, des faiblesses notables subsistent dans les t\u00e2ches n\u00e9cessitant une pens\u00e9e large ou une planification strat\u00e9gique approfondie.<\/p>\n\n\n

Comparaison avec d’autres mod\u00e8les d’IA<\/h3>\n\n\n

Les analyses comparatives directes de Lesswrong et Lennysnewsletter montrent qu’Opus 4.8 surpasse Sonnet 4.6 dans les t\u00e2ches de code et de flux de travail quotidiens \u2013 mais l’\u00e9cart varie selon les d\u00e9tails des t\u00e2ches. GPT-5.5, dont les chiffres cl\u00e9s et les prix manquent, n’a pas encore prouv\u00e9 ses forces dans les grandes \u00e9quipes. Claude Opus 4.8 ne gagne pas parce qu’il est le meilleur dans tout, mais parce qu’il surpasse ses concurrents dans des t\u00e2ches \u00e9troitement d\u00e9finies et r\u00e9p\u00e9tables dont les entreprises ont maintenant besoin. Dans certaines t\u00e2ches de logique de niche, Sonnet 4.6 a encore l’avantage, tandis que GPT-5.5 pourrait avoir une largeur inutilis\u00e9e (bien qu’il ne puisse pas vraiment d\u00e9fier Opus dans le domaine de la programmation sans r\u00e9sultats Pro SWE-bench).<\/p>\n\n\n

Opus 4.8 livre \u2013 mais pas universellement<\/h3>\n\n\n

Opus 4.8 signale un progr\u00e8s constant \u2013 pas une r\u00e9invention radicale \u2013 en se concentrant sur ce qu’il fait d\u00e9j\u00e0 bien : automatisation des processus, g\u00e9n\u00e9ration de code et prototypage rapide pour des \u00e9quipes soucieuses des co\u00fbts. Le mod\u00e8le surpasse sans effort les versions pr\u00e9c\u00e9dentes et les principaux concurrents dans la programmation de masse et les op\u00e9rations routini\u00e8res. Des lancements \u00e9clair, tout en restant abordable et en se d\u00e9veloppant proprement, selon Thezvi","protected":false},"excerpt":{"rendered":"

Claude Opus 4.8 \u00c9valuation : Le dernier mod\u00e8le d’IA d’Anthropic offre des am\u00e9liorations mesurables en mati\u00e8re de code et de flux de travail, mais des sources de premier plan soulignent des faiblesses persistantes dans<\/p>\n","protected":false},"author":5,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_yoast_wpseo_focuskw":"Claude Opus 4.8 Review: Better At What\u2019s It Good At, Worse At What It\u2019s Not","_yoast_wpseo_metadesc":"Claude Opus 4.8 Review: The latest Anthropic AI model delivers measurable code and workflow improvements, but top sources highlight persistent weaknesses in","rank_math_description":"Claude Opus 4.8 Review: The latest Anthropic AI model delivers measurable code and workflow improvements, but top sources highlight persistent weaknesses in","rank_math_focus_keyword":"Claude Opus 4.8 Review: Better At What\u2019s It Good At, Worse At What It\u2019s Not","_stn_article_citation":"","_stn_article_about":"","_stn_article_speakable":"","footnotes":""},"categories":[17,471],"tags":[],"class_list":["post-1379","post","type-post","status-publish","format-standard","hentry","category-ai-crypto","category-ai-crypto-2"],"yoast_head":"\nClaude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faible\u2026<\/title>\n<meta name=\"description\" content=\"Claude Opus 4.8 Review: The latest Anthropic AI model delivers measurable code and workflow improvements, but top sources highlight persistent weaknesses in\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/\" \/>\n<meta property=\"og:locale\" content=\"fr_FR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Claude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faiblesses r\u00e9duites\" \/>\n<meta property=\"og:description\" content=\"Claude Opus 4.8 Review: The latest Anthropic AI model delivers measurable code and workflow improvements, but top sources highlight persistent weaknesses in\" \/>\n<meta property=\"og:url\" content=\"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/\" \/>\n<meta property=\"og:site_name\" content=\"STnews\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/stnews\" \/>\n<meta property=\"article:published_time\" content=\"2026-06-07T13:20:13+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-06-08T14:52:13+00:00\" \/>\n<meta name=\"author\" content=\"Elena Petrova\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:image\" content=\"https:\/\/stnews.live\/app\/uploads\/stn-default-og.png\" \/>\n<meta name=\"twitter:creator\" content=\"@https:\/\/x.com\/epetrova_law\" \/>\n<meta name=\"twitter:site\" content=\"@stnews_live\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Elena Petrova\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"8 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":[\"Organization\",\"NewsMediaOrganization\"],\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/#organization\\\/\",\"name\":\"STnews\",\"alternateName\":[\"ST News\",\"STnews\",\"stnews.live\",\"STnews.live\",\"STN\",\"S T News\",\"\u0421\u0422\u043d\u044c\u044e\u0441\",\"\u30a8\u30b9\u30c6\u30a3\u30fc\u30cb\u30e5\u30fc\u30b9\",\"\u0625\u0633 \u062a\u064a \u0646\u064a\u0648\u0632\"],\"url\":\"https:\\\/\\\/stnews.live\\\/fr\\\/\",\"description\":\"STnews is an independent digital crypto-news platform covering Bitcoin, Ethereum, stablecoins, DeFi, regulatory developments, and on-chain intelligence in 8 languages. Live prices on 250+ coins, editorial analysis, daily briefings, and AI-grounded research.\",\"logo\":{\"@type\":\"ImageObject\",\"url\":\"https:\\\/\\\/stnews.live\\\/fr\\\/wp-content\\\/uploads\\\/2025\\\/05\\\/stnews-logo.png\"},\"sameAs\":[\"https:\\\/\\\/x.com\\\/stnews\",\"https:\\\/\\\/www.linkedin.com\\\/company\\\/stnews\",\"https:\\\/\\\/twitter.com\\\/stnews_live\",\"https:\\\/\\\/x.com\\\/stnews_live\",\"https:\\\/\\\/t.me\\\/stnews_live\",\"https:\\\/\\\/www.youtube.com\\\/@stnews\",\"https:\\\/\\\/github.com\\\/stnews-live\",\"https:\\\/\\\/www.crunchbase.com\\\/organization\\\/stnews\"],\"areaServed\":{\"@type\":\"Place\",\"name\":\"Worldwide\"},\"knowsLanguage\":[\"en\",\"es\",\"ru\",\"tr\",\"uk\",\"ar\",\"fr\",\"pt\"],\"ethicsPolicy\":\"https:\\\/\\\/stnews.live\\\/fr\\\/ethics-policy\\\/\",\"diversityPolicy\":\"https:\\\/\\\/stnews.live\\\/fr\\\/editorial-guidelines\\\/\",\"correctionsPolicy\":\"https:\\\/\\\/stnews.live\\\/fr\\\/corrections\\\/\",\"masthead\":\"https:\\\/\\\/stnews.live\\\/fr\\\/team\\\/\",\"foundingDate\":\"2023-06\",\"publishingPrinciples\":\"https:\\\/\\\/stnews.live\\\/fr\\\/editorial-guidelines\\\/\",\"actionableFeedbackPolicy\":\"https:\\\/\\\/stnews.live\\\/fr\\\/corrections\\\/\",\"verificationFactCheckingPolicy\":\"https:\\\/\\\/stnews.live\\\/fr\\\/editorial-guidelines\\\/\",\"unnamedSourcesPolicy\":\"https:\\\/\\\/stnews.live\\\/fr\\\/editorial-guidelines\\\/\",\"legalName\":\"STnews Media\",\"slogan\":\"Independent crypto news, live prices, and on-chain intelligence.\",\"knowsAbout\":[\"Bitcoin\",\"Ethereum\",\"Cryptocurrency\",\"Blockchain\",\"Stablecoin\",\"Decentralized Finance\",\"Non-fungible token\",\"Web3\",\"Cryptocurrency exchange\",\"On-chain analytics\",\"ETF flows\",\"Halving\",\"Tokenomics\",\"Layer 2 scaling\",\"Spot Bitcoin ETF\"],\"foundingLocation\":{\"@type\":\"Place\",\"name\":\"Tallinn, Estonia\"},\"missionCoveragePrioritiesPolicy\":\"https:\\\/\\\/stnews.live\\\/fr\\\/editorial-guidelines\\\/\",\"ownershipFundingInfo\":\"https:\\\/\\\/stnews.live\\\/fr\\\/about\\\/\"},{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/2026\\\/06\\\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/2026\\\/06\\\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\\\/\"},\"author\":{\"name\":\"Elena Petrova\",\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/#\\\/schema\\\/person\\\/be18d3d32f1a5657d8c428400381db19\"},\"headline\":\"Claude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faiblesses r\u00e9duites\",\"datePublished\":\"2026-06-07T13:20:13+00:00\",\"dateModified\":\"2026-06-08T14:52:13+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/2026\\\/06\\\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\\\/\"},\"wordCount\":1674,\"commentCount\":0,\"articleSection\":[\"AI & Crypto\",\"IA y Cripto\"],\"inLanguage\":\"fr\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/stnews.live\\\/fr\\\/2026\\\/06\\\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/2026\\\/06\\\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\\\/\",\"url\":\"https:\\\/\\\/stnews.live\\\/fr\\\/2026\\\/06\\\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\\\/\",\"name\":\"Claude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faiblesses r\u00e9duites\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/#website\"},\"datePublished\":\"2026-06-07T13:20:13+00:00\",\"dateModified\":\"2026-06-08T14:52:13+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/#\\\/schema\\\/person\\\/be18d3d32f1a5657d8c428400381db19\"},\"description\":\"Claude Opus 4.8 Review: The latest Anthropic AI model delivers measurable code and workflow improvements, but top sources highlight persistent weaknesses in\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/2026\\\/06\\\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\\\/#breadcrumb\"},\"inLanguage\":\"fr\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/stnews.live\\\/fr\\\/2026\\\/06\\\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\\\/\"]}],\"copyrightHolder\":{\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/#organization\\\/\"},\"copyrightYear\":2026,\"publisher\":{\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/#organization\\\/\"}},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/2026\\\/06\\\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/stnews.live\\\/fr\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Claude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faiblesses r\u00e9duites\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/#website\",\"url\":\"https:\\\/\\\/stnews.live\\\/fr\\\/\",\"name\":\"STnews\",\"description\":\"STnews is an independent digital crypto-news platform covering Bitcoin, Ethereum, stablecoins, DeFi, regulatory developments, and on-chain intelligence in 8 languages. Live prices on 250+ coins, editorial analysis, daily briefings, and AI-grounded research.\",\"alternateName\":[\"ST News\",\"STnews\",\"stnews.live\",\"STnews.live\",\"STN\",\"S T News\",\"\u0421\u0422\u043d\u044c\u044e\u0441\",\"\u30a8\u30b9\u30c6\u30a3\u30fc\u30cb\u30e5\u30fc\u30b9\",\"\u0625\u0633 \u062a\u064a \u0646\u064a\u0648\u0632\"],\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/stnews.live\\\/fr\\\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"fr\",\"publisher\":{\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/#organization\\\/\"},\"copyrightYear\":2026,\"copyrightHolder\":{\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/#organization\"}},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/stnews.live\\\/fr\\\/#\\\/schema\\\/person\\\/be18d3d32f1a5657d8c428400381db19\",\"name\":\"Elena Petrova\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr\",\"@id\":\"https:\\\/\\\/stnews.live\\\/app\\\/uploads\\\/2026\\\/05\\\/avatar_elena-petrova.jpg\",\"url\":\"https:\\\/\\\/stnews.live\\\/app\\\/uploads\\\/2026\\\/05\\\/avatar_elena-petrova.jpg\",\"contentUrl\":\"https:\\\/\\\/stnews.live\\\/app\\\/uploads\\\/2026\\\/05\\\/avatar_elena-petrova.jpg\",\"caption\":\"Elena Petrova\"},\"description\":\"Elena Petrova is a regulatory correspondent specializing in crypto law and policy with over 10 years of financial journalism experience. Formerly a finance reporter at Reuters, Elena covers SEC enforcement, MiCA implementation, and global stablecoin regulations. She holds a J.D. from Georgetown Law and is a member of the New York State Bar. Her regulatory analysis is frequently referenced by compliance officers and legal teams at major exchanges.\",\"sameAs\":[\"https:\\\/\\\/www.linkedin.com\\\/in\\\/elenapetrova-law\",\"https:\\\/\\\/x.com\\\/https:\\\/\\\/x.com\\\/epetrova_law\"],\"url\":\"https:\\\/\\\/stnews.live\\\/fr\\\/author\\\/elena-petrova\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Claude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faible\u2026","description":"Claude Opus 4.8 Review: The latest Anthropic AI model delivers measurable code and workflow improvements, but top sources highlight persistent weaknesses in","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/","og_locale":"fr_FR","og_type":"article","og_title":"Claude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faiblesses r\u00e9duites","og_description":"Claude Opus 4.8 Review: The latest Anthropic AI model delivers measurable code and workflow improvements, but top sources highlight persistent weaknesses in","og_url":"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/","og_site_name":"STnews","article_publisher":"https:\/\/www.facebook.com\/stnews","article_published_time":"2026-06-07T13:20:13+00:00","article_modified_time":"2026-06-08T14:52:13+00:00","author":"Elena Petrova","twitter_card":"summary_large_image","twitter_image":"https:\/\/stnews.live\/app\/uploads\/stn-default-og.png","twitter_creator":"@https:\/\/x.com\/epetrova_law","twitter_site":"@stnews_live","twitter_misc":{"Written by":"Elena Petrova","Est. reading time":"8 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":["Organization","NewsMediaOrganization"],"@id":"https:\/\/stnews.live\/fr\/#organization\/","name":"STnews","alternateName":["ST News","STnews","stnews.live","STnews.live","STN","S T News","\u0421\u0422\u043d\u044c\u044e\u0441","\u30a8\u30b9\u30c6\u30a3\u30fc\u30cb\u30e5\u30fc\u30b9","\u0625\u0633 \u062a\u064a \u0646\u064a\u0648\u0632"],"url":"https:\/\/stnews.live\/fr\/","description":"STnews is an independent digital crypto-news platform covering Bitcoin, Ethereum, stablecoins, DeFi, regulatory developments, and on-chain intelligence in 8 languages. Live prices on 250+ coins, editorial analysis, daily briefings, and AI-grounded research.","logo":{"@type":"ImageObject","url":"https:\/\/stnews.live\/fr\/wp-content\/uploads\/2025\/05\/stnews-logo.png"},"sameAs":["https:\/\/x.com\/stnews","https:\/\/www.linkedin.com\/company\/stnews","https:\/\/twitter.com\/stnews_live","https:\/\/x.com\/stnews_live","https:\/\/t.me\/stnews_live","https:\/\/www.youtube.com\/@stnews","https:\/\/github.com\/stnews-live","https:\/\/www.crunchbase.com\/organization\/stnews"],"areaServed":{"@type":"Place","name":"Worldwide"},"knowsLanguage":["en","es","ru","tr","uk","ar","fr","pt"],"ethicsPolicy":"https:\/\/stnews.live\/fr\/ethics-policy\/","diversityPolicy":"https:\/\/stnews.live\/fr\/editorial-guidelines\/","correctionsPolicy":"https:\/\/stnews.live\/fr\/corrections\/","masthead":"https:\/\/stnews.live\/fr\/team\/","foundingDate":"2023-06","publishingPrinciples":"https:\/\/stnews.live\/fr\/editorial-guidelines\/","actionableFeedbackPolicy":"https:\/\/stnews.live\/fr\/corrections\/","verificationFactCheckingPolicy":"https:\/\/stnews.live\/fr\/editorial-guidelines\/","unnamedSourcesPolicy":"https:\/\/stnews.live\/fr\/editorial-guidelines\/","legalName":"STnews Media","slogan":"Independent crypto news, live prices, and on-chain intelligence.","knowsAbout":["Bitcoin","Ethereum","Cryptocurrency","Blockchain","Stablecoin","Decentralized Finance","Non-fungible token","Web3","Cryptocurrency exchange","On-chain analytics","ETF flows","Halving","Tokenomics","Layer 2 scaling","Spot Bitcoin ETF"],"foundingLocation":{"@type":"Place","name":"Tallinn, Estonia"},"missionCoveragePrioritiesPolicy":"https:\/\/stnews.live\/fr\/editorial-guidelines\/","ownershipFundingInfo":"https:\/\/stnews.live\/fr\/about\/"},{"@type":"NewsArticle","@id":"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/#article","isPartOf":{"@id":"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/"},"author":{"name":"Elena Petrova","@id":"https:\/\/stnews.live\/fr\/#\/schema\/person\/be18d3d32f1a5657d8c428400381db19"},"headline":"Claude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faiblesses r\u00e9duites","datePublished":"2026-06-07T13:20:13+00:00","dateModified":"2026-06-08T14:52:13+00:00","mainEntityOfPage":{"@id":"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/"},"wordCount":1674,"commentCount":0,"articleSection":["AI & Crypto","IA y Cripto"],"inLanguage":"fr","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/","url":"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/","name":"Claude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faiblesses r\u00e9duites","isPartOf":{"@id":"https:\/\/stnews.live\/fr\/#website"},"datePublished":"2026-06-07T13:20:13+00:00","dateModified":"2026-06-08T14:52:13+00:00","author":{"@id":"https:\/\/stnews.live\/fr\/#\/schema\/person\/be18d3d32f1a5657d8c428400381db19"},"description":"Claude Opus 4.8 Review: The latest Anthropic AI model delivers measurable code and workflow improvements, but top sources highlight persistent weaknesses in","breadcrumb":{"@id":"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/#breadcrumb"},"inLanguage":"fr","potentialAction":[{"@type":"ReadAction","target":["https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/"]}],"copyrightHolder":{"@id":"https:\/\/stnews.live\/fr\/#organization\/"},"copyrightYear":2026,"publisher":{"@id":"https:\/\/stnews.live\/fr\/#organization\/"}},{"@type":"BreadcrumbList","@id":"https:\/\/stnews.live\/fr\/2026\/06\/claude-opus-48-evaluation-improved-strengths-diminished-weaknesses-6\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/stnews.live\/fr\/"},{"@type":"ListItem","position":2,"name":"Claude Opus 4.8 \u00c9valuation : Forces am\u00e9lior\u00e9es, faiblesses r\u00e9duites"}]},{"@type":"WebSite","@id":"https:\/\/stnews.live\/fr\/#website","url":"https:\/\/stnews.live\/fr\/","name":"STnews","description":"STnews is an independent digital crypto-news platform covering Bitcoin, Ethereum, stablecoins, DeFi, regulatory developments, and on-chain intelligence in 8 languages. Live prices on 250+ coins, editorial analysis, daily briefings, and AI-grounded research.","alternateName":["ST News","STnews","stnews.live","STnews.live","STN","S T News","\u0421\u0422\u043d\u044c\u044e\u0441","\u30a8\u30b9\u30c6\u30a3\u30fc\u30cb\u30e5\u30fc\u30b9","\u0625\u0633 \u062a\u064a \u0646\u064a\u0648\u0632"],"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/stnews.live\/fr\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"fr","publisher":{"@id":"https:\/\/stnews.live\/fr\/#organization\/"},"copyrightYear":2026,"copyrightHolder":{"@id":"https:\/\/stnews.live\/fr\/#organization"}},{"@type":"Person","@id":"https:\/\/stnews.live\/fr\/#\/schema\/person\/be18d3d32f1a5657d8c428400381db19","name":"Elena Petrova","image":{"@type":"ImageObject","inLanguage":"fr","@id":"https:\/\/stnews.live\/app\/uploads\/2026\/05\/avatar_elena-petrova.jpg","url":"https:\/\/stnews.live\/app\/uploads\/2026\/05\/avatar_elena-petrova.jpg","contentUrl":"https:\/\/stnews.live\/app\/uploads\/2026\/05\/avatar_elena-petrova.jpg","caption":"Elena Petrova"},"description":"Elena Petrova is a regulatory correspondent specializing in crypto law and policy with over 10 years of financial journalism experience. Formerly a finance reporter at Reuters, Elena covers SEC enforcement, MiCA implementation, and global stablecoin regulations. She holds a J.D. from Georgetown Law and is a member of the New York State Bar. Her regulatory analysis is frequently referenced by compliance officers and legal teams at major exchanges.","sameAs":["https:\/\/www.linkedin.com\/in\/elenapetrova-law","https:\/\/x.com\/https:\/\/x.com\/epetrova_law"],"url":"https:\/\/stnews.live\/fr\/author\/elena-petrova\/"}]}},"_links":{"self":[{"href":"https:\/\/stnews.live\/fr\/wp-json\/wp\/v2\/posts\/1379","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/stnews.live\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/stnews.live\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/stnews.live\/fr\/wp-json\/wp\/v2\/users\/5"}],"replies":[{"embeddable":true,"href":"https:\/\/stnews.live\/fr\/wp-json\/wp\/v2\/comments?post=1379"}],"version-history":[{"count":1,"href":"https:\/\/stnews.live\/fr\/wp-json\/wp\/v2\/posts\/1379\/revisions"}],"predecessor-version":[{"id":1418,"href":"https:\/\/stnews.live\/fr\/wp-json\/wp\/v2\/posts\/1379\/revisions\/1418"}],"wp:attachment":[{"href":"https:\/\/stnews.live\/fr\/wp-json\/wp\/v2\/media?parent=1379"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/stnews.live\/fr\/wp-json\/wp\/v2\/categories?post=1379"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/stnews.live\/fr\/wp-json\/wp\/v2\/tags?post=1379"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}