Annonces
L'actualité tech en une phrase — le fait le plus marquant de chaque article publié.
Binance Pay reports 200% transaction growth in Argentina as the service expands to seven countries with national QR code integration and fiat balance payment su
Cursor launches a ₹649/month India plan and an Auto model router to cut agent costs, plus full PR review on iPhone and iPad; Start arrives July 28, 2026.
France's SNIA strategy targets 100,000 AI students per year, 15% of the embedded AI market, and 400 SMEs equipped by 2025, backed by 2.5 billion euros in public
Silent Hill: Townfall launches September 24 on PS5 with first-person combat, stealth-focused gameplay, and a story that blurs reality and hallucination.
Audio8 releases a CPU-only ONNX Runtime for Audio8-TTS-Preview-0.6B-ONNX-INT4, with INT4 weights, streaming PCM, and voice cloning in about 1 GiB of RAM.
Microsoft announces Project Perception, an agentic security system with multi-model AI defense, entering public preview on August 3.
dtControl2+ε extends dtControl2 with a tunable epsilon parameter, pruning decision trees for MDP controllers by sacrificing a controlled amount of optimality fo
Alibaba's ARMS Node.js agent bundles OTel tracing, runtime metrics, and AI observability into one npm package for Node.js BFFs and AI orchestration.
BenchLM ranks Grok 4.1 Fast the top voice-agent LLM at 0.54s first answer, while Claude Opus 4.6, its highest scorer, takes 1.78s and misses the call-quality ba
Coinbase relaunches token sales in the U.S. with a 'filling up from the bottom' allocation algorithm, accepting USDC requests, and penalizing flippers with smal
Reve released Reve 2.1, claiming second place on Arena while training on less than a tenth of the compute of the labs around it, with 4K output.
OpenAI invested in Reve; part of Reve's AI research team joins OpenAI to lead multimodal research, while Reve keeps its products and independence.
Mistral released Shieldstral, a 3B safety classifier that takes moderation policies as plain-language questions at inference, matching guardrails 7x its size.
Shieldstral, a 3B policy-adaptive multimodal safety classifier, matches text classifiers nearly seven times its size and sets a new multimodal SOTA, per arXiv.
Liquid AI released open-weight LFM2.5-Encoders (230M and 350M) for CPU-friendly long-context NLP, claiming 3.7× faster inference than ModernBERT-base.
GPU costs accrue by the calendar hour, so utilization, not model quality, is becoming AI's binding constraint, and few teams own keeping clusters busy.
Tiny Roar's Lou's Lagoon launches on PS5 August 27, featuring customizable characters, seaplanes, and homes, a Deluxe Edition with exclusive items, and PS5 hapt
Alibaba's Qoder 1.0 gives every agent task isolated worktree boundaries so tasks don't collide, cutting input tokens 40% in its internal memory A/B test.
A research paper proposes AgentToolMO, a standardized 3GPP NRM information model for cross-vendor agent tool trust management, to contain cascading failures in
NHTSA opened a preliminary probe into suspension failures in nearly 1.2 million Tesla Model 3 and Model Y vehicles after 156 complaints of a part detaching with