MiniMax Launches M2.5 Open-Source AI for Coding and Tasks
The Chinese AI lab trained M2.5 with reinforcement learning across 200,000 real-world coding environments in over 10 languages, enabling it to plan projects like an architect before coding full development cycles. It scores 80.2% on SWE-Bench Verified—edging Opus 4.6's 78.9%—runs 37% faster than its predecessor, and delivers at 100 tokens per second for just $1 an hour.
Anthropic Raises $30 Billion at $380 Billion Valuation
The Series G round, led by GIC and Coatue with big names like Microsoft and NVIDIA joining in, will support research, product innovation, and scaling Claude for enterprises. The company reported $14 billion in run-rate revenue, driven by tools like Claude Code which powers 4% of global GitHub public commits and serves eight of the top 10 Fortune 500 companies.
OpenAI released GPT-5.3-Codex-Spark to ChatGPT Pro subscribers through the Codex app, command-line, and VS Code extension.
This smaller, speed-focused version of GPT-5.3-Codex generates over 1,000 tokens per second—up to 15 times faster on coding benchmarks—thanks to Cerebras' Wafer Scale Engine 3 chip. The tool aims to act as a seamless pair programmer, with plans for broader rollouts and multimodal features ahead.