AI digest: Meta goes closed source, OpenAI cuts prices
Meta releases its first closed-source frontier model while OpenAI halves Pro pricing and Anthropic limits cybersecurity capabilities.
Big week for model releases and business strategy shifts. Meta’s going proprietary, OpenAI’s getting competitive on pricing, and everyone’s worried about AI finding security holes.
Meta ships Muse Spark without open weights
Meta Superintelligence Labs released Muse Spark, their first frontier model and notably their first without open weights. The multimodal reasoning model supports tool use, visual chain of thought, and multi-agent orchestration. This marks a major strategy shift from Meta’s previous open approach, likely driven by competitive pressure as they chase OpenAI and Anthropic’s capabilities.
OpenAI launches $100 Pro plan to fill pricing gap
OpenAI introduced a $100 monthly Pro tier aimed at heavy Codex users, sitting between the $20 Plus and $200 Pro plans. The move directly undercuts Anthropic and Google’s pricing whilst giving developers more Codex access at half the previous Pro cost. Smart positioning as coding becomes the killer use case for these models.
Anthropic keeps Mythos locked down over security fears
Anthropic limited release of its Claude Mythos model after it found thousands of cybersecurity vulnerabilities across major operating systems and browsers. They’re quietly sharing it with responsible organisations through “Project Glasswing” instead of public release. Whether this is genuine safety concern or competitive advantage masquerading as responsibility remains an open question.
NVIDIA releases KVPress for memory-efficient inference
NVIDIA’s KVPress tackles the memory bottleneck in long-context LLM inference through KV cache compression. The practical coding guide shows how to implement memory-efficient generation for extended conversations. This matters because memory, not compute, is often the limiting factor for deploying these models at scale.