All AI news

AISGLang says restructured MLA decode kernels on Rubin, which fit a deeper pipeline in 327 KiB of shared memory, deliver a 16% speedup at batch 16 with 128K context and bit-identical output. The post also reports 20% faster full FP8 MLA at batch 1 and 20% faster KDA verify kernels after keeping weights in registers and reducing synchronization. It additionally covers fusing MoE finalization, the shared expert, 8-GPU all-reduce, and RMSNorm into one collective kernel.

AIMiles runs the full reinforcement learning loop on Nvidia Rubin, using SGLang for rollout, Megatron for training, and one container image. On a single 4-GPU tray, Qwen3-30B-A3B's GSM8K reward rises from about 45% to about 95% over 50 rollouts, matching the GB300 curve. The post also reports DeepSeek-V4-Flash end-to-end rollout and training, and Qwen3.5-35B-A3B agentic RL with 64 concurrent mini-SWE-agent sandboxes on SWE-bench Verified, where reward holds near 0.6 and median response length falls about 30%.

AISGLang says it worked with NVIDIA to optimize attention, MoE, and speculative verification kernels for Kimi K3 inference on early-access Rubin hardware. It reports up to 20% faster FP8 MLA at batch 1 with 128K context, 20% faster KDA verification with bitwise-identical output, and a 5.9% end-to-end speedup from MoE tail fusion that removes 276 kernel launches per decode step. The post also says SGLang powers Miles' end-to-end RL training on Rubin, including agentic RL with 64 concurrent sandboxes on the Vera CPU.

AIPrime Intellect says it rewrote its Prime Agent coding tool in Rust, using a swarm of more than 2,000 agents over two weeks. The company reports cold start to typing about 13 times faster than the TypeScript version, and memory use over 80% lower after startup. Prime Agent remains open source and adds native Windows support in beta and Homebrew installation.

Why it matters: The post shows how a multi-agent swarm rewrote a coding agent with parity checks, giving a concrete case of agent-driven software engineering with measured results.

AIA GitHub post announces that kernel kits for DeepSeek's TP=2 and TP=4 recipes are now live. The post links to a repository named deepseek-v4.1-tensorfold-tp2-2xgb10 and gives no further details on performance or features.

AIbig-arrow-on-the-screen (bigarrow) is a MIT-licensed macOS command-line tool and skill for Claude Code and Codex that draws arrows, boxes and text over any window. Clicks pass through, keyboard focus stays put, and each arrow removes itself after a set duration or when its agent process ends. The tool only points; it never clicks, types or captures the screen, and it requires no macOS permission to draw.

AIPrime Intellect says it is turning the workflow behind a recent rewrite into reusable state machines in Prime Agent, letting users run their own agent teams through implementation, review, and verification. The company also says it is accelerating work on capabilities and evals and connecting Prime Agent with cloud agent swarms and hosted training for autonomous research. This release adds native Windows support in beta and Homebrew installation.

AIPrime Intellect says Prime Agent is now open source and that its Rust rewrite is available today. The full blog post covers verification loops, multi-agent orchestration, and performance experiments.

AIPrime Intellect says users can update their Prime Agent by running the command prime-agent update. The post links to the project's GitHub repository, PrimeIntellect-ai/prime-agent.

AIDatology's team tested its Curation Studio by generating large volumes of synthetic data on Modal, scaling to hundreds of B300 GPUs on demand. The post frames Curation Studio as a customizable data curation tool for teams training their own models.

AIDex Horthy (@dexhorthy) posted only the words "he cook" on X, with no further detail. The post is a short reaction and does not describe a product, release, or result on its own. Background from the quoted post by @0xblacklight describes a serverless background agent that created a GitHub pull request from an issue.

AIPrime Intellect says swarms have costs, since messages consume tokens, lose information, and agents must coordinate to avoid duplicated work. The company is extending its RL training infrastructure from individual agents to multi-agent systems, letting developers express arbitrary agent interactions and train them.

AIGoodfire says its activation monitors use signals from inside a model to detect undesired behaviors, catching more cases, running faster and costing less than text-based monitors. Baseten customers can use them to monitor for prompt injection, actions outside policy, sensitive data exposure and cyber misuse.

AIGoodfire says teams can configure how their applications respond when a concern is flagged, including logging, additional review, refusal, and re-routing. The company directs model servers and trainers to a partnership post with Baseten for building monitors into their stack.

AIRadixArk says Proximal is using Miles to train coding agents and calls it a flexible, scalable foundation for teams running their own training workloads. Proximal says its training framework is built on Miles, with runs on Modal's on-demand GPU clusters and serverless GPUs for inference. Its sandboxing infrastructure runs on Kubernetes and can handle millions of concurrent rollouts.

AICline is offering Solar Mini 4 free, a new 35B mixture-of-experts model from Korean lab Upstage with 3B active parameters. It has a 524K context window and runs at 208 tokens per second. Cline says it scores 24 on the AAII, the highest of any model at 3B active and within one point of Nemotron 3 Ultra, which uses 55B active.

AIMagnific promotes Brand Kit in Magnific One for building campaigns. The post says Magnific Desktop is unlimited for Brand Kit users until October 15.

AIBoardroom is a free, open-source tool that lets users hold real-time meetings with their AI agents instead of relaying messages between them. The creator says it launches today on Product Hunt.

AIAgentzon announces a storefront built for AI agents, drawing on two years of work on agentic commerce products and tools. The post frames the product as a response to what agents want, without listing features, prices, or availability. Background from Paul Graham says Amazon's ban on agents creates an opening for startups to compete with Amazon.

AIAgentzon launches at agentzon.co as a checkout site for AI assistants, aggregating about 500 household essentials so far. Its founder says most AI assistant checkouts are poor because Amazon blocks assistants from shopping, and users can ask an agent to buy specific brands such as Colgate toothpaste. The site supports browser-use or API checkout, WebMCP, UCP and OpenAPI, and requires no accounts or API keys.

AIArtificial Analysis ran 30 consecutive real estate staging edits through four image editing models, and Ideogram 4.5 kept most of each image unchanged while others drifted. Ideogram 4.5 and FLUX 3 left 95% or more of the image untouched on small edits, while GPT Image 2.5 Sunburst re-rendered most of the image and left only about a fifth unchanged. Nano Banana 2.1 kept its edits local but gradually darkened the rest of the image.