Erdős Proof Lands Amid Compute Squeeze
- OpenAI's internal reasoning model delivered a verifiable proof overturning Erdős's planar unit-distance conjecture, confirmed by mathematicians
-...

Created by tao hong
Latest Manus AI funding, product updates, technical research, tutorials, and competitive market analysis
Explore the latest content tracked by Manus AI Radar
Anthropic's $30B round at $900B+ valuation marks a sharp shift in AI power dynamics.
The infrastructure layer for AI agents is advancing from basic execution to resilient, integrated platforms.
Alibaba's Qwen3.7-Max is generating buzz for topping LM Arena rankings and launching just a month after its predecessor, yet real-world tests reveal...
Manus co-founders are racing to raise ~$1B to unwind Meta's $2B acquisition after China's NDRC blocked it in April.
Hermes Agent OS provides a free toolkit for automating lead outreach, content calendars, and customer support.
Developers building reliable AI agents must move beyond basic prompts to layered engineering approaches.
China's state-backed investors are pouring capital into the country's leading AI startups, with a sharp focus on large language model development. This signals deepening official support for domestic AI champions.
Google is giving away its fastest model that beats the Pro flagship on five benchmarks, signaling a clear split between accessible speed layers and...
Butterfly Effect built Manus as an autonomous AI agent but remains ensnared by Chinese regulations even after incorporating abroad. Oversight now resembles a regulatory windfall tax more than routine national security review.
Anthropic's first projected profitable quarter hinges on a nonrecurring compute discount from SpaceX that may exceed the $559M figure, despite a $1.25B monthly commitment through May 2029 with reduced fees only in May-June 2026.
Manus founders are exploring a $1 billion raise from outside investors specifically to buy the company back from Meta and complete an unwind. This strategy highlights their push for full independence in the competitive AI landscape.
Google's I/O 2026 unveiling of AI agents as a new consumer web interface stood out as both promising and the most confusing announcement. The mixed reception highlights real doubts about whether everyday users will embrace the ecosystem.
Google's Gemini 3.5 Flash has taken the #1 spot on Zapier's Automation Bench, outperforming all other frontier models in agentic tasks at a fraction of the cost. This underscores its strong edge in practical automation workflows.
Converting agent traces into SFT datasets is positioned as the future of agent training, with the teich library offering a simple pip install to enable this workflow for better data preparation.
Model providers get the headlines, but runtime, identity, data, and payments decide whether agents reach production.
Dari-docs uses parallel coding agents to optimize documentation, as featured on Hacker News with 16 points. This offers a practical example of agent tooling for streamlined content workflows.