Scaling Codex to enterprises worldwide
OpenAI launches Codex Labs and partners with Accenture, PwC, and Infosys to help enterprises deploy Codex across the software lifecycle. The initiative aims to scale Codex adoption and has reached 4M Codex WAU.
Le meilleur du dev et de l'IA, scoré chaque jour par un agent. Filtré, résumé, classé. Aucune couleur, aucun bruit — juste la matière.
OpenAI launches Codex Labs and partners with Accenture, PwC, and Infosys to help enterprises deploy Codex across the software lifecycle. The initiative aims to scale Codex adoption and has reached 4M Codex WAU.
L’article analyse comment les systèmes autonomes combinant modèles LLM et outils open source permettent de détecter, corriger et propager rapidement des patches de vulnérabilités. Il défend l’ouverture comme avantage structurel face aux risques liés aux boîtes noires et aux reverse-engineering assistés par l’IA. Les systèmes embarqués comptent plus que la seule taille des modèles.
llm-openrouter 0.6 adds a refresh command to refresh the list of available models without waiting for the cache to expire, enabling quick testing of Kimi 2.6 on OpenRouter. The post notes a Pelican-generated HTML page for Kimi, including an HTML/JavaScript UI to control the animation, plus a transcript.
95% of cancer treatments fail in clinical trials, and Noetik frames this as a matching problem between patients and therapies. They propose autoregressive transformers, like TARIO-2, to model this matching and improve trial success rates.
Session timeouts, if mishandled, disrupt essential tasks and create accessibility barriers for users with disabilities. The article argues for thoughtful session management that balances security with usability, offering strategies like visible warnings, non-destructive re-authentication, and configurable timeout behavior.
Kimi K2.6 from Moonshot AI is now available on Vercel AI Gateway. It targets long-horizon coding tasks across Rust, Go, and Python, and can turn prompts into complete front-end interfaces while improving autonomous agents’ stability and safety. AI Gateway offers a unified API for model calls, usage/cost tracking, retries, and observability, with Bring Your Own Key and leaderboard access.
The post experiments with pulling data from a Datasette instance into Google Sheets. It outlines three approaches: using importdata(), a named function wrapping it, and a Google Apps Script for API tokens. An example sheet demonstrates all three.
The Claude Token Counter now allows comparing token counts across models (Opus 4.7/4.6, Sonnet 4.6, Haiku 4.5). Opus 4.7 changes the tokenizer, inflating token counts (up to ~1.46x vs 4.6) and pricing; it also improves image support. The post shares benchmarks for text, images, and PDFs to help estimate costs.
Headless services are poised to become mainstream for personal AIs, delivering faster, more reliable experiences than GUI-driven bots. Salesforce's Headless 360 demonstrates APIs-as-UI, exposing Salesforce, Slack, and related platforms as APIs, which could upend per-head SaaS pricing.
L'article détaille l'évolution du prompt système entre Claude Opus 4.6 et 4.7, avec des changements majeurs : renommage de la plateforme, ajout d'outils (Claude in Powerpoint), extension des instructions safety enfants, et mécanisme de recherche d'outils. Il insiste sur le ton moins assertif et des réponses plus concises.
Claude system prompts are turned into a git timeline: Anthropic's system prompt history is converted into a granular, timestamped commit history. This enables tracing evolution with git log, diff, and blame by organizing prompts per model, family, and revision.
Vercel ne supprime plus le dernier déploiement de prévisualisation des branches avec des PR ouvertes ou non fusionnées dans les politiques de rétention. Les fenêtres de rétention plus courtes restent sûres pour les déploiements actifs. Les 10 déploiements de production les plus récents et les alias restent préservés, quelles que soient les règles.
Zo Computer used Vercel's AI SDK and AI Gateway to replace per-provider adapters and complex routing. The result was a 20x drop in retry rate, 99.93% chat success, and substantial latency reductions, enabling rapid model support with a config-based approach.
Anthropic Claude Opus 4.7 is touted as the next SOTA model, claiming improvements over 4.6 across dimensions. The release frames it as a stronger all-around performer for LLM tasks. The brief note signals rapid benchmarking and evaluation to follow.
Vercel Flags est désormais disponible en GA: un fournisseur de feature flags intégré à la plateforme Vercel. Il permet de créer et gérer des flags avec ciblage, segments d'utilisateurs et contrôles d'environnement directement dans le dashboard, et propose un SDK Flags pour Next.js et SvelteKit. Pour les autres frameworks ou backends, l'adaptateur Vercel Flags supporte OpenFeature pour s’intégrer via un SDK indépendant.
Claude Opus 4.7 is now available on Vercel AI Gateway, optimized for long-running, asynchronous agents with improved memory and stronger image-processing for charts and documents. It introduces task budgets to manage token use and configurable thinking content, while AI Gateway provides a unified API, usage tracking and automatic retries.
Vercel's Workflows introduces a durable execution model where orchestration lives in your application code, removing the need for a separate scheduler. It offers isolated steps, retries, durable state, and streaming, plus deep AI/agent integration and built-in encryption. Available in TypeScript and Python via the Workflow SDK, with Worlds and Fluid compute for deployment and execution.
Leading security firms and enterprises join OpenAI's Trusted Access for Cyber, leveraging GPT-5.4-Cyber and $10M in API grants to reinforce global cyber defense. The move signals a coordinated effort to provide secure, scalable cyber tooling to security teams and vendors worldwide.
L’article montre comment entraîner ou finetuner des modèles d’embeddings et de reranker multimodaux (texte, images, audio, vidéo) avec Sentence Transformers. Il détaille les composants d’entraînement (modèle, dataset, loss, arguments, évaluateur, trainer) à travers un cas pratique de Visual Document Retrieval avec Qwen3-VL-Embedding-2B, améliorant significativement le NDCG@10.
L’article propose un Skill et un test harness pour porter des modèles transformers vers MLX-lm, en s’appuyant sur des agents code. Il explique comment concevoir des contributions de haute qualité dans un écosystème où les agents génèrent beaucoup de PR peu fiables, tout en augmentant le signal pour les reviewers.