Google Research open-sources RRSI, letting agents improve prompts, tools and memory without overfitting, lifting ...

NVIDIA's Open Agent Safety Platform pairs OpenShell sandboxes with BlueField-4 Sentry, quarantining AI agents that escape in ...

Alibaba Qwen-Audio-3.1-Realtime is a full-duplex voice agent model with 262K context, tool calling, web search and GRPO training.

Fireworks AI's Ember-1 post-trains Kimi K3 for shorter reasoning, using about 40% fewer tokens while holding benchmark ...

Google Research's AI video co-director uses 4 agentic frameworks on Gemini and Veo to generate coherent, minutes-long videos.

Supersonic Labs' Julia 1 is a 144.3M-parameter Apache 2.0 decision model that classifies, scores, and routes on CPU.

TypeSafe AI's Jev turns agent decisions into typed probabilities. See 20 agentic use cases, benchmarks, pricing, and limits.

Sarvam AI's Saaras V4 speech-to-text covers 22 Indian languages, adds keyterm prompting, 5 output modes, sub-150 ms streaming ...

Black Forest Labs (BFL), the lab behind the FLUX image models, has released FLUX 3 Action. It is a 7B open-weights World ...

GitHub Copilot, AWS Kiro, Cursor and Cognition compared on IP indemnity, prompt storage, audit logs and 500-seat cost.

Anthropic redesigned Claude Code Projects: a coordinator splits work into parallel cloud threads, beta on Pro and Max.

OpenAI launches GPT-6 Sol and Luna, lower-cost models priced from $0.10 per million input tokens with caching upgrades.