Six AI engineering signals from the last 24 hours:
Codex hits 7M users after 10x growth in six months.
Anthropic's tokenizer inflates costs versus competing models.
A Claude API budget cap reportedly overran its stated limit by 7x.
Apple's planned M7 Ultra could support up to 1.5 TB of unified memory.
A Microsoft study examines early Claude Code and Copilot CLI adoption.
A new paper applies ablation and statistical validation to KV-cache compression.
Tokenizer pricing, runtime budget enforcement, and hardware memory ceilings are all shifting at once. Cost models locked in six months ago are now stale at every layer of the inference stack.
[link] [comments]