Anthropic released Claude 4.6 Opus in March with a 1 million token context window (first Opus-class model to achieve this), 75.6% on SWE-bench coding tasks, and 128K output tokens. The model's frontier performance on complex reasoning tasks and its expanded context capability make it the technical leader for ambiguous multi-file codebases and knowledge work.

Google released Gemini 3.1 Pro with zero price increase ($2/$12 per million tokens) while achieving a major reasoning performance leap: 77.1% on ARC-AGI-2 (more than double predecessor) and 1M token context window. This release demonstrates Google's strategy to compete through efficiency improvements and rapid iteration rather than price increases.

Anthropic's Claude Sonnet 4.6 debuted in March with a 1 million token context window in beta and maintains unchanged $3/$15 pricing, making it the new default free model on claude.ai. User surveys show 59% prefer Sonnet 4.6 over Opus 4.5, indicating that efficiency and cost improvements matter more than raw capability increments in production deployments.

OpenAI announced its sixth acquisition of 2026 on March 19 (Astral for open-source tools), nearly matching its entire 2025 acquisition count. This aggressive M&A strategy reflects OpenAI's $110 billion funding round and billion-dollar ARR, positioning the company to rapidly build competitive capabilities and maintain market dominance.

Defense AI company Shield AI announced on March 26 a $1.5 billion Series G at $12.7B valuation (doubling from prior $5.6B) and a $500 million fixed-return Blackstone deal. The company is acquiring Aechelon, a defense simulation specialist, to integrate high-fidelity simulation with its Hivemind autonomy platform for military applications.

Databricks announced a confidential IPO filing on March 31 for Q2 2026 roadshow, targeting $105-110B valuation. With $4.5B ARR, 12,500+ enterprise customers, and DBRX 2.1 open model gaining traction, the company positions itself as the enterprise data and AI platform at the center of the AI stack.

Cursor's Composer 2 model release ignited transparency controversy when it emerged that 25% of the model derived from Kimi K2.5 base, with Cursor initially downplaying attribution. The incident highlights geopolitical tensions in the AI stack and raises questions about open-source model interdependence between US and Chinese AI ecosystems.

Google DeepMind announced a strategic partnership with Munich-based Agile Robots on March 24 to integrate Gemini Robotics foundation models into Agile Robots' hardware across manufacturing, automotive, and logistics. With over 20,000 deployed robotic systems globally, this partnership enables real-world deployment data collection to improve AI models for physical tasks.