Nvidia announced RTX Spark at Computex, a new chip class for Windows PCs delivering 1 petaflop of AI compute and 128GB unified memory, purpose-built to run local agents.

• RTX Spark PCs from Microsoft, Dell, and others will ship this fall; Jensen Huang called the shift as significant as the smartphone reinvention of the phone.
• Nvidia also announced Vera Rubin entering full production, with CoreWeave claiming the first validated NVL72 bring-up and the platform delivering 10x lower inference cost per token versus Grace-Blackwell.

MiniMax released M3 today, an open-weight model scoring 59% on SWE-Bench Pro with a one-million-token context window and a sparse attention architecture that cuts compute to one-twentieth of dense baselines.

• MiniMax Sparse Attention speeds up input processing by more than 9x and the model benchmarks on par with Opus 4.7 and GPT-5.5 on long-horizon autonomy tests.
• Weights will be published shortly after API availability; the model also supports native multimodality and computer use.

Anthropic launched dynamic workflows in Claude Code today in research preview, allowing a single session to write its own orchestration scripts and spin up tens to hundreds of parallel subagents.

• Available now in Claude Code CLI, Desktop, and VS Code extension for Max, Team, and Enterprise plan users.
• The feature targets engineering tasks too large for a single agent pass, coordinating subagents that divide work, cross-check findings, and return a single result without requiring users to build a pipeline.

GitHub replaced Copilot's flat-rate subscription with usage-based AI Credits billing effective June 1, with code completions remaining free but every chat, agentic session, and code review consuming credits at $0.01 each.

• Early cost projections from developers show bills tracking from $29 flat-rate subscriptions toward $750 or more, with no automatic downgrade fallback to cheaper models when credits run out.
• Business and Enterprise plans receive a temporary credit boost during transition, but the removal of the safety-net fallback is the structural change that will force procurement renegotiations.

CMA-CGM Group began rolling out MAIA, Powered by Mistral, to approximately 80,000 employees across CMA-CGM, Ceva Logistics, and CMA Media starting June 1.

• The deployment covers the full group workforce across three distinct business units, making it one of the largest single-vendor agentic AI rollouts in logistics.
• CMA-CGM signed a five-year strategic partnership with Mistral in April 2025, and today marks the beginning of the phased production rollout rather than a pilot.

BCG disclosed that its sales AI agent Jamie is trained on losing calls alongside winning ones, generating post-call scorecards for salespeople and real-time guidance for AI agents during live conversations.

• The dual-training approach targets identification of what drives customers away, not just what closes deals, addressing a known gap in AI sales coaching that typically optimizes only on positive outcomes.
• Real-time agent guidance during live calls represents a production deployment of inference-time coaching, distinct from asynchronous post-call analysis.