An AI just cracked 80-year-old math — plus Qwen 3.7, Gemini Omni, Cursor Composer 2.5
An OpenAI reasoning model autonomously disproved Erdős's 1946 unit-distance conjecture. Alibaba's Qwen 3.7 Max became the highest-ranked Chinese model in the world. Google's Gemini Omni shipped video generation with conversational editing. Cursor Composer 2.5 matched Claude Opus 4.7 at one-tenth the cost. Four stories. One week.
Google I/O 2026 — Gemini Spark, Antigravity 2.0, Search agents, Universal Cart, XR glasses. Everything Google shipped today.
Gemini 3.5 Flash beat 3.1 Pro at 4x speed. Gemini Omni for video. Gemini Spark — a $100/mo personal agent. Antigravity 2.0 demoed building Doom on stage. AI Mode hit 1B MAU. Universal Cart for shopping. AI Inbox. Android XR glasses. The complete list, with opinion.
Seedance 2.0 just invented N64 games that never existed. And a full energy drink ad.
ByteDance's Seedance 2.0 + GPT Image 2 are making creators ship complete video commercials and fake-retro-game footage in hours.
AI video generation just moved to China — and Western media barely noticed
Sora is dead, Grok briefly took the crown, and the top three blind-test rankings are now Chinese.
OpenAI shipped GPT-5.5-Cyber to vetted security teams. Sam Altman just did exactly what he criticized Anthropic for.
Restricted access to a powerful cyber model. Vulnerability ID, patch validation, malware analysis. The same playbook Altman called 'fear-based marketing' a month ago.
OpenAI just shipped 3 production voice models: gpt-realtime-2, Translate, and Whisper
Voice agents finally have GPT-5-class reasoning, multi-tool calling, and live translation across 70+ languages. The pricing tells you who they're built for.
OpenAI ships GPT-5.5: what actually changes for builders
The model is faster, cheaper per token, and finally usable for long-running agent loops. Here's what the launch post buries — and what it means for anyone shipping production AI.
Llama 4 405B is out: open weights at frontier-adjacent quality
Meta released Llama 4 with weights you can actually run. The 405B variant is GPT-5-class on most evals, the 70B is shockingly good, and the licensing terms have one important catch.
Google ships Gemini 3 with native tool use — quietly
Gemini 3 lands without the usual launch theatrics. The interesting story isn't the eval scores — it's that tool calling is now a first-class part of the model, not a wrapper bolted on top.
- An AI just cracked 80-year-old math — plus Qwen 3.7, Gemini Omni, Cursor Composer 2.5
An OpenAI reasoning model autonomously disproved Erdős's 1946 unit-distance conjecture. Alibaba's Qwen 3.7 Max became the highest-ranked Chinese model in the world. Google's Gemini Omni shipped video generation with conversational editing. Cursor Composer 2.5 matched Claude Opus 4.7 at one-tenth the cost. Four stories. One week.
- Google I/O 2026 — Gemini Spark, Antigravity 2.0, Search agents, Universal Cart, XR glasses. Everything Google shipped today.
Gemini 3.5 Flash beat 3.1 Pro at 4x speed. Gemini Omni for video. Gemini Spark — a $100/mo personal agent. Antigravity 2.0 demoed building Doom on stage. AI Mode hit 1B MAU. Universal Cart for shopping. AI Inbox. Android XR glasses. The complete list, with opinion.
- Seedance 2.0 just invented N64 games that never existed. And a full energy drink ad.
ByteDance's Seedance 2.0 + GPT Image 2 are making creators ship complete video commercials and fake-retro-game footage in hours.
- AI video generation just moved to China — and Western media barely noticed
Sora is dead, Grok briefly took the crown, and the top three blind-test rankings are now Chinese.
- OpenAI shipped GPT-5.5-Cyber to vetted security teams. Sam Altman just did exactly what he criticized Anthropic for.
Restricted access to a powerful cyber model. Vulnerability ID, patch validation, malware analysis. The same playbook Altman called 'fear-based marketing' a month ago.
- OpenAI just shipped 3 production voice models: gpt-realtime-2, Translate, and Whisper
Voice agents finally have GPT-5-class reasoning, multi-tool calling, and live translation across 70+ languages. The pricing tells you who they're built for.
- OpenAI ships GPT-5.5: what actually changes for builders
The model is faster, cheaper per token, and finally usable for long-running agent loops. Here's what the launch post buries — and what it means for anyone shipping production AI.
- Llama 4 405B is out: open weights at frontier-adjacent quality
Meta released Llama 4 with weights you can actually run. The 405B variant is GPT-5-class on most evals, the 70B is shockingly good, and the licensing terms have one important catch.
- Google ships Gemini 3 with native tool use — quietly
Gemini 3 lands without the usual launch theatrics. The interesting story isn't the eval scores — it's that tool calling is now a first-class part of the model, not a wrapper bolted on top.
