All articles in the news category.
A hackathon project proves small models can power complex multi-agent systems through design, not scale. Five AI traders on Qwen2.5-3B create emergent market crashes, wealth gaps, and price swings—without a single call to GPT-4.
Cursor restructured its pricing this week, cutting annual Teams costs by 20% while rolling out enterprise governance tools for budget control. The moves come as the AI coding industry abandons flat-rate subscriptions in favor of consumption-based billing.
NVIDIA released Nemotron 3.5 Content Safety, a 4B-parameter model that combines multimodal input evaluation, multilingual reach across 140 languages, custom enterprise policy enforcement, and auditable reasoning traces in one inference call. The model addresses critical gaps in production AI safety pipelines.
Nvidia released Nemotron 3 Ultra, a 550-billion-parameter open-weight model optimized for long-running agents. While it's the fastest among U.S. open-weight models and promises 30% cost savings, it still lags behind Chinese competitors and GPT-5.5 on core benchmarks.
Global IT services firm Endava has embedded AI agents throughout its software development process, from requirements gathering to deployment. The move signals a shift from AI as a coding assistant to AI as an autonomous workflow participant.
At Build 2026, Microsoft doubled down on a contrarian thesis: enterprise AI needs organizational memory more than bigger models. The company launched HorizonDB, GPU-accelerated warehousing, and made Fabric IQ generally available to give agents the context layer they're missing.
Anthropic is scaling Project Glasswing to 150 new organizations, providing access to Claude Mythos Preview for vulnerability detection. The expansion comes amid concerns about validation transparency and the race to secure critical infrastructure before offensive AI capabilities proliferate.
OpenAI is repositioning Codex beyond developers, adding Sites for shareable interactive dashboards, extended Annotations for documents, and curated plugins for sales, finance, and legal teams. With 1 million knowledge workers already using the platform weekly, this marks a direct challenge to Anthropic's Claude Cowork.
GitHub officially switched Copilot from flat-rate subscriptions to usage-based billing tied to token consumption. While plan prices stay the same, heavy users are reporting dramatic cost increases as model choice now directly impacts spending.
H Company's Holo3.1 brings computer-use agents to consumer hardware with quantized checkpoints and mobile support. The release includes four model sizes and delivers 79.3% accuracy on AndroidWorld while running entirely locally.
SkipLabs just launched Skipper, an AI coding agent that rejects the prompt-review-iterate cycle entirely. Instead of making developers faster, it aims to make their involvement optional—generating complete, validated backend services from a single prompt.
NVIDIA ships Cosmos 3, an omni-model that unifies world generation, physical reasoning, and action prediction in a single architecture. The release marks a shift from separate specialized models to one foundation model for robotics, autonomous vehicles, and smart spaces.
Replit is embedding Visa's payment infrastructure directly into its development platform, giving AI agents a cryptographic identity layer and native transaction capabilities. The partnership signals a shift from bolting payments onto finished products to building commerce into agents from day one.
Google AI Studio's new Antigravity coding agent lets non-developers build functional applications through natural language prompts. A Google editor with no coding background built a working I/O 2026 quiz to demonstrate the capability.
Google unveiled Gemini Omni, a multimodal model that generates and edits video through natural language, alongside Gemini 3.5 Flash, designed for complex agentic workflows. Both models are rolling out to consumers and developers with significant implications for content creation and enterprise automation.
Snyk entered the AI pentesting market with Evo Continuous Offensive Security, targeting the 350-day gap left by traditional security testing. The platform uses LLM reasoning for context-dependent flaws while reserving deterministic scanning for known vulnerability classes.
Anthropic released Claude Opus 4.8 with user-controlled effort levels, parallel subagents for large coding tasks, and fast mode at one-third the previous cost. The model also shows significant improvements in honesty and reduced deception rates.
Snowflake is betting big on AI with a $6 billion, five-year commitment to AWS for compute and GPU resources. Under CEO Sridhar Ramaswamy, the data warehouse company is repositioning itself as an AI platform, leveraging cost-efficient Graviton processors to subsidize expensive model training workloads.
Tokenmaxxing—treating AI token usage as a productivity metric—is draining enterprise budgets. Uber's CTO admitted their Anthropic Claude budget exploded. New tools like Lanai Token Tuner aim to shift focus from token gluttony to measurable business outcomes.
Google I/O 2026's Dialogues stage brought together CEO Sundar Pichai, DeepMind's Demis Hassabis, and quantum computing experts to discuss proactive AI agents, quantum-AI convergence, and AI's expanding role in science and creativity. The sessions signal Google's push beyond chatbots into autonomous agents and quantum-accelerated AI research.
Google's new Gemini 3.5 Flash model is matching or beating flagship models from OpenAI and Anthropic in several benchmarks—while delivering tokens 4x faster at a fraction of the cost. The performance gap between 'fast' and 'frontier' models is closing.
xAI launches Grok 4 with unprecedented reasoning capabilities, real-time web access, and multimodal understanding that puts it in direct competition with the latest from OpenAI and Anthropic.
AWS now offers direct access to Anthropic's Claude Platform using AWS credentials, but there's a critical data residency catch. Here's what developers need to know about this new integration versus using Claude on Amazon Bedrock.
While Elon Musk battles Sam Altman in federal court, he just rented his entire Memphis AI supercomputer to Anthropic. The deal exposes how compute capacity—not model performance—now defines competitive advantage in frontier AI.
OpenAI just released a Chrome extension that connects Codex directly into your browser, allowing agents to work across authenticated sessions and multiple tabs without commandeering your desktop. This moves AI agents closer to where modern work actually happens.
OpenAI just released GPT-Realtime-2, bringing GPT-5-level reasoning to voice interactions with a 4x larger context window. The update includes two specialized models for translation and transcription, signaling a push toward voice-first AI applications.
OpenAI released GPT-5.5 Instant, an incremental update focusing on conversational quality, reasoning accuracy, and user personalization. The model targets real-world deployment needs rather than raw benchmark gains.
OpenAI has released Symphony, an open-source specification designed to standardize how AI agents coordinate and communicate. The move signals a strategic shift toward interoperability in multi-agent systems, potentially reshaping how developers build complex AI workflows.
OpenAI has released GPT-5.5, marking a significant iteration in large language model capabilities. The model introduces enhanced multimodal reasoning, extended context windows, and improved computational efficiency without requiring architectural overhaul.
OpenAI has confirmed a security incident involving compromised developer credentials from the Axios HTTP library. The breach affects a subset of ChatGPT users, marking a concerning reminder of supply chain vulnerabilities in AI infrastructure.
OpenAI has opened research preview access to Codex Security, an AI system designed to automatically detect security vulnerabilities in code. The tool extends OpenAI's Codex technology into application security, targeting a market desperate for automated vulnerability detection.
OpenAI has released GPT-5.4, the first significant update to its GPT-5 architecture. The model promises enhanced reasoning capabilities and reduced response times without the infrastructure overhaul that marked the GPT-4 to GPT-5 transition.
OpenAI just removed the paywall between free users and its most powerful publicly available model. ChatGPT free tier now includes GPT-4o access, Advanced Voice Mode, and canvas editing tools—features previously locked behind the $20/month Plus subscription.
OpenAI has introduced Lockdown Mode and Elevated Risk labels to ChatGPT, marking a significant shift in how the platform handles security for journalists, activists, and other high-risk users. The new features aim to protect against sophisticated prompt injection and social engineering attacks.
OpenAI's GPT-5 has achieved a breakthrough in biotechnology by reducing the cost of cell-free protein synthesis through AI-optimized reaction conditions. The development could accelerate drug discovery and make synthetic biology more accessible.
OpenAI has released GPT-5.3-Codex, a code-focused model designed to outperform GPT-4 on programming tasks. The company claims 90% accuracy on the HumanEval benchmark and positions it as a direct competitor to GitHub Copilot and Anthropic's Claude for code generation.
OpenAI and Snowflake announced a strategic partnership bringing GPT models directly into Snowflake's data cloud platform. The integration eliminates data movement requirements and enables enterprises to deploy frontier AI on their existing infrastructure.
OpenAI and SoftBank have enlisted SB Energy to deliver renewable power for the Stargate AI infrastructure project. The partnership addresses the massive energy demands of advanced AI data centers with a focus on clean energy sources.
