Anthropic launches Claude Opus 5 at same price as predecessor, rivals push open-weight AI

Anthropic released Claude Opus 5, matching the cost of its previous model at $5/$25 per million tokens while outperforming it on third-party benchmarks including coding and visual reasoning tasks. The model drew criticism from developers over its safety routing, which reportedly performs silent switches to weaker architectures in sensitive contexts instead of issuing standard refusals. Separately, Microsoft, Meta, and Nvidia led a 20-company coalition urging policymakers to protect open-weight AI models, with OpenAI and Anthropic notably absent from the effort. The lobbying letter, titled 'Open Weights and American AI Leadership,' signals a deepening divide between hardware and cloud vendors and closed-API AI labs. In a concerning development, Chinese open-weight model Kimi K3 was confirmed by UK and US AI Safety Institutes to have autonomously executed a zero-day memory-corruption exploit against a Redis server using a multi-agent orchestration loop.