AI News

OpenAI launches GPT-5.6 Max with a 10M token context window — 10x larger than GPT-5.6 Sol. The new model targets enterprise agent workloads requiring entire codebase comprehension, with $0.50/M input tokens and native tool-calling support.

The EU AI Act Phase 2 enforcement begins today: 40% of enterprise AI agents are now classified as high-risk, requiring audit trails, human oversight gates, and real-time monitoring. Non-compliance fines reach 7% of global revenue.

Hugging Face released Open-Agent Protocol 1.0 today, an open-source agent communication standard backed by a Linux Foundation working group. OAP 1.0 combines MCP tool discovery, A2A agent-to-agent messaging, and a new Agent Card v2 format with capability negotiation.

Microsoft released Azure Agent Fabric today, a managed platform for enterprise multi-agent orchestration with built-in role-based access control, cost budgets, audit trails, and agent-to-agent protocol support. The platform targets Fortune 500 companies deploying agent fleets at scale.

Google released Gemini 4.0 Flash today with a 10M token context window, native MCP-compatible tool calling, and $0.15/M input token pricing. The model processes hour-long videos and 1,000-page documents in a single pass, eliminating the chunking and retrieval overhead that has defined RAG architecture since 2023.

OpenAI drops GPT-5.6 Nano at $0.10 per million tokens — 250x cheaper than Sol. The 3B parameter model runs on consumer GPUs, targets the factual lookup and simple workflow tier, and signals OpenAI's push to dominate every layer of the agent cost stack.

Anthropic releases Claude Code 2.0 with a 100K file context window and full codebase rewriting. The terminal coding agent can now refactor entire repositories, generate cross-file changes, and create pull requests with full test coverage in a single session.

NVIDIA unveils Vera Rubin, a next-gen GPU architecture delivering 4x inference throughput for AI agent workloads. With 512GB HBM4 memory and 2x interconnect bandwidth, it targets the inference bottleneck that currently limits agent fleet scaling.

NVIDIA has unveiled the Blackwell Ultra B300, featuring 288GB HBM4 memory, 2x inference throughput over the B200, and a new interconnect architecture that processes 10M plus token contexts in a single GPU pass.

OpenAI has launched GPT-5.6 Turbo, delivering 3x faster inference than GPT-5.6 Sol at 50% lower cost, while matching Sol's quality on 94% of benchmarks. The release eliminates the speed-quality tradeoff and positions Turbo as the default model for latency-sensitive agent workloads.

Anthropic has launched Claude 5 Enterprise, its most capable model yet with a 2M token context window, native tool execution, and enterprise SSO — arriving just as the company confirms crossing $2B ARR. The release positions Anthropic as the enterprise AI leader against OpenAI's GPT-5.6 Sol and Google's Gemini 3.7 Pro.

Anthropic has formally updated its internal risk assessment, elevating catastrophic-misalignment risk to 'low' while disclosing a powerful, unreleased 'Model 2' held back for safety.