Agentic Inference Capacity Grew 4.6x in a Year
AI chips sold over the past year added 4.6 times the agentic inference capacity, while peak compute grew only 2.6 times. Blackwell and better design drove the gap.
AI chips sold over the past year added 4.6 times the agentic inference capacity, while peak compute grew only 2.6 times. Blackwell and better design drove the gap.
Google released Gemini 4 Neon, priced at $3 and $15 per million input and output tokens with a 1 million-token context window, aimed at coding and agent work.
Google launched Genie 4 in Project Genie for eligible AI Ultra subscribers, with mid-session prompts, native audio, a world-building agent and a community gallery.
Newsletter Funda AI argues that Claude Code drove Anthropic’s growth to 50% a month, but rising training costs now make AI revenue harder to forecast.
Epoch AI says Anthropic compute spending is rising about 4x a year against 3x for OpenAI, with $447.5 billion already committed through 2030.
A Hugging Face post shows MoE expert streaming running a 180B model on a 32GB Mac at roughly 14 tokens a second, within 6% of Strata.
OpenAI has unveiled GPT-6.2 Astra and Sol, two advanced models designed to boost productivity for professional users.
A Substack essay says a crude form of recursive self-improvement is under way, citing METR’s finding that AI R&D at Anthropic has roughly doubled in speed.