Google's Gemini 3.5 Flash Beats Its Own Pro at 4x the Speed
Gemini 3.5 Flash turns images into interactive code, beats last-gen Pro on agents and coding, at 4x the speed
20 stories
Gemini 3.5 Flash turns images into interactive code, beats last-gen Pro on agents and coding, at 4x the speed
Cohere open-sources North Mini Code, a 30B MoE coding model with 3B active params, built for agentic software engineering and on-prem deployment
LMSYS's Miles framework introduces Token-In-Token-Out (TITO), a design principle that eliminates silent training drift in multi-turn agentic RL by guaranteeing bit-perfect token consistency across rollout turns.

Cohere's North Mini Code is a 30B MoE coding agent that fits on one H100, runs free under Apache 2.0, and beats models four times its size on coding benchmarks.
Google and Hugging Face launch a live, multi-agent competition to push Gemma 4 E4B inference speed as high as possible on a single A10G GPU — with results updating in real time.

Google's Gemini 3.5 Live Translate streams speech across 70+ languages in near real-time, and is now open to developers via the Gemini Live API

Google's Gemini 3.5 Live Translate drops the old STT-MT-TTS pipeline for a single end-to-end audio model covering 70+ languages at $0.023/min

Google's new audio model streams speech translation across 70+ languages in near real-time, collapsing the classic transcribe-translate-synthesize pipeline into a single model call

Google's Gemini 3.5 Live Translate streams speech across 70+ languages in near real-time, preserving voice, pitch, and pacing — and it's available now via API, Google Translate, and Google Meet.
Google's Gemini 3.5 Live Translate delivers near real-time speech-to-speech translation across 70+ languages while preserving tone, pace, and pitch

Luma's Ray3.2 brings 16-keyframe control, 8-face tracking, 20-second 1080p clips, and a first-ever API to professional video pipelines
Higgsfield's seven AI tools now live natively inside DaVinci Resolve — generate footage, remove backgrounds, match color grades, and upscale to 8K without leaving your timeline
Arcee AI signs a multi-million-dollar deal making Hugging Face the exclusive home for every model, dataset, and agent trace it builds

Kimi Code v0.12.0 ships /swarm for parallel agent execution, native proxy support, and graduates three experimental features to stable defaults
Tencent's UniRL lets one RL training loop post-train diffusion models, LLMs, VLMs, and hybrid architectures — with two new algorithms included

Unsloth packages BAAI's bge-small-en-v1.5 in GGUF format, making a 33M-param embedding model runnable anywhere llama.cpp runs — no GPU required.

Fish Audio open-sources S2 Pro, a 4B-parameter dual-autoregressive TTS system trained on 10M hours that beats closed-source rivals on Seed-TTS eval.
Perplexity brings its AI startup competition to the UK, offering £1M in Computer credits to teams who can build a unicorn in 12 weeks
Z-Image-Engineer V6 is a fine-tuned text encoder that rewrites simple prompts into cinematic descriptions and slots directly into Z-Image-Turbo workflows
Google brings Gemini cloud models into Apple's Foundation Models framework and Xcode 27, letting iOS/macOS developers swap on-device and cloud inference with a single API.