Google's Antigravity 2.0 Builds a Working OS With 93 Parallel AI Agents
Google's Antigravity 2.0 turns a single prompt into a playable arcade game on physical hardware using AI subagents, live at Google I/O 2026
20 stories
Google's Antigravity 2.0 turns a single prompt into a playable arcade game on physical hardware using AI subagents, live at Google I/O 2026
A new theoretical framework uses Hopfield energy to predict which samples diffusion models forget first — and which to replay to prevent it
EAGLE 3.1 fixes a newly discovered 'attention drift' flaw in speculative decoding, delivering up to 2x faster long-context throughput natively in vLLM
OpenBMB's MiniCPM5-1B tops the 1B open-weights leaderboard with a 17.9 Intelligence Index score, beating 2B models while fitting in 0.5GB
Epoch AI models every Blackwell GPU on Earth and finds token demand growing 10x/year while supply grows just 3-4x — a crunch may already be here
Warp's agent now accepts any OpenRouter-compatible URL, giving developers instant access to 300+ models with a single alias command
Figure AI's humanoid robots are heading to JCPenney's supply chain, marking one of the first real commercial deployments of general-purpose humanoids in retail logistics
Google I/O 2026 drops Gemini 3.5 Flash, Antigravity 2.0, and a full agentic dev stack — all generally available today

vLLM's new Rust API server hits 5x throughput on preprocessing-heavy workloads, enabled by a single environment variable

Microsoft's MAI-Image-2.5 debuts at No. 3 on Arena's text-to-image leaderboard, with a +75-point jump over its predecessor and the biggest gains in text rendering and commercial imagery.

PrismML's Bonsai Image 4B squeezes a 4-billion-parameter diffusion model to under 1 GB, making on-device image generation on iPhone a reality for the first time.

Gradium's 100M-parameter on-device TTS model hits 1.00% WER on Seed-TTS, beating models up to 5x its size while adding int8 quantization and removing latency-killing input padding.

Alibaba's Qwen3.7-Max arrives with a 1M-token context window, top-tier agentic benchmarks, and a 35-hour autonomous coding run that doubles its predecessor's output

Inception is hiring its first Forward Deployed AI Engineer as enterprise demand for Mercury 2 — the world's only production diffusion LLM hitting 1,000+ tokens/sec — outpaces what its research team can handle alone.
Warp's AI agent now accepts any OpenAI-compatible endpoint, letting you swap in OpenRouter, LiteLLM, or a private gateway and switch models with a slash command.

An open-source Zig binary indexes your repo into in-memory structures and exposes 21 MCP tools so coding agents skip the ripgrep spawn.

World Labs and OpenArt show how a single image can become a persistent, navigable 3D set that creators direct like a film shoot

Lovable now ships a built-in analytics dashboard with every published app, giving builders real-time visitor data, traffic sources, and device breakdowns with zero setup.

Meta's open-source TLX Block Attention kernel delivers 2.3x faster attention and 3.5x fused backward speedup on NVIDIA Blackwell GPUs by exploiting compile-time knowledge of block-diagonal attention patterns.

Runway's Project Luxo claims AI video has crossed the uncanny valley, with a solo creator building a 10-minute film in under a month that moved Hollywood insiders to tears.