Grok Voice 3.0 achieves a score of 86.9 on the Artificial Analysis Speech-to-Speech Index, ranking #1 among tested models.
The open-weight reasoning model features 30B parameters with 3B active per token and a 1M token context window.
Administrators can now set daily quotas on AI input and output tokens at both the project and per-user levels to prevent budget overruns.
The method achieves a mean F1 score of 0.798 across 14 language pairs in CS-FLEURS using the Whisper-small encoder.
The study presents the first tri-modal Masked Diffusion Models pretrained on text, image-text, and audio-text data, achieving strong results at 3B scale.
AI21 analysis of internal traffic reveals public coding benchmarks overestimate task difficulty, leading to inefficient model routing.
Codex and ChatGPT Work now account for more than half of output tokens among EU ChatGPT Enterprise customers.
Improves SSM kernel performance by up to 1.8x at batch size 256 and speculative decoding throughput by up to 2.1x at batch size 512
DeepSeek V4.1 Flash now accounts for 22.6% of all tokens routed through OpenRouter at $1.20 per million output tokens.
The update includes 79 PRs and 8 direct commits from 97 contributors, focusing on fixing blocked updates and preventing data loss during repairs.
Semi-OPD outperforms on-policy distillation in 14 of 17 tested teacher-student pairs, achieving up to 11.4x training speedup.
The new open-source Parsimony benchmark quantifies AI coding efficiency by counting net syntax units added, revealing that Gemini 3.1 Pro Preview adds the least code among top performers.
OpenAI used its own AI models to identify the employee who planted the device, leading to an FBI arrest.
Gemini 4 Argon reduces prompt injection success rates to 0.7% and cuts coding task cheating attempts to 2% through new multi-layered defenses.
The new pipeline reduces median post-push scan execution time by 86% and improves user acceptance by 11.5% compared to the previous LLM approach.
The update introduces a playground tab for testing decision models like Kev and OpenJev with probability bars.
The program connects systems integrators and consultancies with organizations to implement Replit's AI-powered software development tools.
The open-weight model achieves 184 tokens/s on an Apple M5 Max and scores 48.09 on LiveClaw, outperforming larger models in agentic tasks.
The framework achieves a KL divergence of 1.01 against real user queries, a 9.2x improvement over previous baselines.
The platform is deployed across 120,000 Orange employees and adopted by over 150 customers in France and Europe.
Cloudflare D1 surpassed 5.5 million databases by decoupling control plane operations from query execution using Workers KV and Durable Objects.
Gavin Baker argues that lower model-layer margins will drive compute demand, citing a 37% IRR for B200 clusters even with steep price declines.
The new model improves conversion rates by 18.9% and click-through rates by 12% across Meta's advertising surfaces.
The new model offers introductory pricing of $2 per million input tokens and $10 per million output tokens, with a 1 million token context limit.
AI Briefing