01Text
Mercury 2
Fastest reasoning LLM, diffusion architecture with 1,009 tokens per second and four times lower prices
ArchiveWeek 09/2026 · 1 March 2026
Text, design, data / knowledge, research and agents: one tool each, checked for value, cost and data protection.
Week 9/2026 marks a break in AI development: speed, accessibility and orchestration take centre stage. The standout event was the launch of Mercury 2 by Inception Labs, the first diffusion-based reasoning model, which opens up a new speed dimension at more than 1,000 tokens per second. QuiverAI entered the stage with Arrow-1, the first dedicated AI model for vector graphics, which immediately topped the SVG Arena leaderboard and raised 8.3 million dollars from a16z. Perplexity Computer redefines the AI agent by orchestrating 19 different models into one autonomous workflow system. Notion Custom Agents brings 24/7 AI automation into one of the most widely used workspaces, without programming, and Supaboard shows that data analysis no longer requires expert knowledge when you can simply ask in English.
This issue dates from before the English edition launched in October 2026. Summaries and short descriptions are translated; the detailed reviews are available in German. Detailed review in German →
01Text
Fastest reasoning LLM, diffusion architecture with 1,009 tokens per second and four times lower prices
02Design
First AI model for vector graphics: editable SVGs from text prompts, number one on SVG Arena
03Data / Knowledge
Ask your data in plain English: #1 on Product Hunt in February 2026, 600+ integrations, no SQL needed
04Research
19 AI models orchestrated: an autonomous multi-model agent for complex workflows, 200 dollars per month
05Agents
24/7 autonomous AI agents directly in Notion, free until May 2026, no coding required
The detailed version with all research notes is available in German.