
Moonshot AI's Kimi-K2.7-Code Slashes Reasoning Costs by 30% on 1T Open-Weight Model
Moonshot AI open-sources Kimi-K2.7-Code, a 1T-parameter coding agent that cuts reasoning token usage 30% while beating K2.6 on every benchmark
20 stories

Moonshot AI open-sources Kimi-K2.7-Code, a 1T-parameter coding agent that cuts reasoning token usage 30% while beating K2.6 on every benchmark
Claude Fable 5's safety fallback system caused Artificial Analysis's Intelligence Frontier chart to decline for the first time in its history, raising hard questions about what 'frontier' really means.
The US Commerce Department invoked national security export controls to force Anthropic to pull its most powerful AI models from all users worldwide, citing a reported jailbreak.
Artificial Analysis launches the first hardware benchmark built for agentic workloads, using real coding agent trajectories and production-grade optimizations to measure Agents per Megawatt
Browser Use's stealth cloud browser is now a one-package install inside Anthropic's Claude Managed Agents, giving hosted agents live web access with CAPTCHA solving and residential proxies at $0.02/hr.
Google opens Gemini 3.5 Live Translate to developers via the Live API, enabling continuous speech-to-speech translation across 70+ languages with just a few lines of config

Moonshot AI's Kimi K2.7 Code brings 30% fewer reasoning tokens and major coding benchmark gains over K2.6, now open-source and live in Kimi Code CLI
serve-sim streams any booted iOS Simulator to the browser over MJPEG, giving AI agents like Codex and Cursor a live, interactive window into your app with one npx command
SGLang breaks 12,000 tokens/GPU/second on DeepSeek V4 Pro using NVIDIA's GB300 NVL72 rack, Dynamo orchestration, and MTP speculative decoding

Claude Fable 5 hits 88% on FrontierMath's hardest tier, up from under 10% just months ago — on a freshly error-corrected benchmark
Replit Agent 4 lets you run parallel AI agents to simultaneously build a web app, mobile app, video, and pitch deck from a single project
MiniMax M3, the first open-weight model combining frontier coding, 1M-token context, and native multimodality, is now available inside Factory's Droid agent platform
Google Research publishes two studies showing AI skin condition tools triple laypeople's diagnostic accuracy, but struggle to guide next steps

Moonshot AI's Kimi K2.7 Code brings a 1T-parameter open-weight coding agent with 30% fewer reasoning tokens and forced multi-turn thinking, targeting long-horizon agentic workflows.
GitHub Copilot code review now pulls context from MCP servers and custom agent skills, letting teams enforce their own standards on every PR automatically.
Epoch AI overhauled FrontierMath after finding errors in 42% of problems, and the corrected scores reveal AI math ability is advancing faster than the benchmark itself
Google's new audio-to-audio model streams speech translation across 70+ languages in seconds, killing the transcribe-translate-synthesize pipeline for good

Google ships five AI launches in one week: real-time voice translation, a turbocharged NotebookLM, global Project Genie access, expanded Gemini Notebooks, and an experimental text-diffusion open model.
Google Research's GeminiSQL-2 tops the BIRD benchmark for text-to-SQL, using a fine-tuned Gemini 3.1 Pro to turn plain English into execution-ready queries
World Labs releases three research papers turning single images and videos into full 3D and 4D scenes using generative diffusion models