Open source

AIRadixArk says Miles runs reinforcement learning end to end on NVIDIA Vera Rubin, using SGLang rollouts, Megatron training and one container image. Agentic RL runs 64 concurrent sandboxes on the Vera CPU next to the GPUs. The linked SGLang post reports that Kimi K3 inference gained up to 20% faster FP8 MLA at batch 1 with 128K context, and a 5.9% end-to-end speedup from MoE tail fusion.

Why it matters: The post gives concrete speedup figures and a specific RL setup on early-access Rubin hardware, useful for engineers comparing inference and training stacks.

AIvLLM now supports NVIDIA Vera Rubin NVL72, with daily container builds and support for models from DeepSeek, Moonshot AI, Z.ai, and MiniMax. In early AgentX benchmarks, vLLM running MiniMax M3 delivered up to 7.84x the throughput per GPU of GB200 NVL72 at matched interactivity. The post is an early look, and the team expects further gains from ongoing optimizations.

Why it matters: The post gives specific hardware specs, kernel configurations, and benchmark figures for running vLLM on Vera Rubin NVL72, useful for teams planning deployments.

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.

AISierra has published a draft of the Personal Agent Protocol, known as Poppy, and named 35 additional design partners, including Adyen, Bank of America, Mastercard, OpenAI, PayPal, and Visa. Under the protocol, companies publish a /.well-known/poppy.json discovery file, and personal agents start sessions, identify themselves, and sign in through OAuth with session tokens limited to approved access. The company says the draft will be followed by design workshops and a reference implementation over the next month.

Why it matters: The draft specifies how personal agents identify themselves, obtain customer-approved access, and work with company websites, APIs, or agents, which helps readers assess its practical effect on agent-driven transactions.

AISherwin Wu, an OpenAI employee, says the updated Harvey LAB-AA v1.1 benchmark, announced by Artificial Analysis with Harvey, is more useful than the original LAB results. The new Hallucination-Gated All-Pass Rate credits a task only when every rubric criterion passes and no material hallucination appears. Grok 4.7 (xhigh) leads at 9.4%, while GPT-6 Astra (max) at 8.6% has very few material hallucinations.

Why it matters: The update adds a hallucination gate to a legal benchmark, showing that models with high all-pass rates can rank much lower once material errors count.

AINVIDIA Dynamo uses a unified session-level identifier to make its inference stack aware of agent sessions, subagents, and their KV cache across turns and tool calls. On SWE-bench, two TP4 MiniMax-M2 replicas on one 8xH100 node gained roughly 12-16% throughput from program-aware scheduling over KV-aware routing alone. The post also describes experimental shared-pool indexing and a proposed KvHint interface for session-aware cache policies in vLLM and SGLang.

Why it matters: The post explains how session identifiers let an inference stack track agent working sets, with measured throughput gains on SWE-bench and agentic RL rollouts.

AICarbon-A is an open model that predicts gene locations directly from DNA, and it has been used to annotate genomes from over 22,000 species. The release includes a database of 566 million gene candidates, about 16 times the gene annotations in the RefSeq dataset. Wet-lab RNA experiments supported 239 candidates missing from RefSeq across cats, Syrian hamsters, chickens, and Arabidopsis.

Why it matters: The source ties an open gene-annotation model to specific wet-lab checks and gene counts, helping readers judge how far its predictions extend beyond well-studied genomes.

AIJetBrains released Mellum2.1, a 12B mixture-of-experts model with 2.5B active parameters under the Apache 2.0 license, built for coding agents. Post-training shifted to reinforcement learning across thousands of environments and millions of sandboxed runs, and the model is available on Hugging Face. The source reports gains over Mellum2 on LiveCodeBench, AIME, GPQA Diamond, BFCL v4, IFEval, and SWE-bench Verified, and says it serves almost twice the tokens of Qwen3.5-9B under heavy load.

Why it matters: The post shows how reinforcement learning in real sandboxed environments changed a compact open model's repository work, with benchmark gains against Mellum2 and two peers.

AIAnthropic has launched the Anthropic Cyber Mission, which starts with the Critical Infrastructure Defense Program for operational technology and OSS Scanner for open-source projects. The defense program brings frontier Claude models, on-site engineers and threat research to trusted providers such as Accenture, CrowdStrike and Palo Alto Networks. OSS Scanner gives enrolled open-source projects periodic free scans from its strongest models, with reports sent without human review and an expected true-positive rate above 90%.

Why it matters: The announcement shows how a frontier AI lab is packaging cyber defense around critical infrastructure and open-source maintainers, including the program's partners and access routes.

AIOpenAI published 722 math manuscripts covering 372 result groups in a new GitHub repository, openai/math, all produced by an unreleased internal model. The author describes the results as including a near-Riemann hypothesis claim pushed to 0.875, and notes that 25 Fields Medal winners criticized the company's approach to AI math research.

Why it matters: The piece traces how AI math results moved from benchmarks to open problems, offering context on verification and the mathematicians' pushback.

AIEpoch gave six models 11 real work tasks from its own operations, including graphic design, data insights, and research design, and graded outputs against employee standards. Fable 5.1 and GPT-6 Astra led on average task performance, reliably handling well-defined work such as coding and computational analysis. The report finds that all models still fail on open-ended judgment, including matching Epoch's standards, designing informative experiments, and generating diverse ideas, so the authors conclude AI cannot yet replace workers at Epoch.

Why it matters: The report separates well-defined task reliability from open-ended judgment failures, which benchmark scores on easily verifiable tasks would miss.

AIGoogle's AI Edge Team open-sourced ML Drift under Apache 2.0, a GPU compute engine for on-device AI inference across OpenGL ES, OpenCL, Metal, and WebGPU. It serves as the core GPU acceleration engine within LiteRT and succeeds the legacy TFLite GPU delegate, which will no longer receive new features. The post cites benchmarks showing up to 40% lower frame latency in YouTube Shorts and up to 30% faster on-device performance in Adobe Lightroom and Photoshop.

Why it matters: The post explains how ML Drift unifies GPU shaders across platforms and replaces the TFLite GPU delegate, which matters for developers deploying on-device models.

AIA Hugging Face blog author used the ML-Intern agent in HuggingChat to build six small models by writing detailed prompts that specify datasets, base models, baselines, smoke tests, and spending limits. The projects include a citrus disease vision-language model, a Huggy character LoRA, a camera-angle LoRA, a doodle-to-object LoRA, a 0.8B prompt rewriter, and a 4-step distilled Agate model, with total compute cost of about USD 103. Each project's prompts and public models are linked from the post.

Why it matters: The author shows how prompt structure, baselines, smoke tests, and budget caps shape an agent-driven training workflow, with per-project costs given.

AIPerplexity is open-sourcing pplx-embed-v2-late, multi-vector embedding models for text and images in one shared space, in 9B and 0.6B sizes. The 9B model can index multimodal data, the 0.6B model can run queries on device, and PDF pages can be searched without OCR. The author reports 92.4% on MADQA and 64% on BrowseComp+, with weights available on Hugging Face.

Why it matters: Two open-weight multi-vector models share one space for text and images, with a 0.6B on-device option, a useful comparison for building multimodal retrieval.

AIMicrosoft Research Asia has open-sourced Agent Lightning v1.0, a roughly 3,500-line agentic RL framework that trains the same agent harness used in deployment. In an end-to-end coding agent pipeline, Qwen3.5-9B rose from 41.8% to 56.4% Pass@1 on SWE-bench Verified using about 6,000 training samples. The framework runs agents as standard Kubernetes jobs without paid commercial sandbox services.

Why it matters: The source shows how training with the deployed agent harness avoids rebuilding agents, and reports concrete SWE-bench Verified gains from about 6,000 samples.

AINVIDIA reports that fine-tuned Nemotron models reached gold-medal level at both IOI 2026, scoring 535.4 out of 600, and IMO 2026, scoring 30 out of 42. The IOI run was a live, unofficial, unsupervised benchmark, while IMO proofs were graded by official IMO graders. The post also releases checkpoints, datasets, a new 200-problem benchmark, and inference pipelines on Hugging Face and NeMo-Skills.

Why it matters: The post traces how SFT, RL, and a generate-verify-refine loop turned Nemotron into gold-level specialists for IOI and IMO, with the training and inference details shared.

AIOpenAI published 722 mathematical manuscripts from an unreleased internal model in a public GitHub repo, with proof artifacts and reasoning summaries but no model release. The source says the results are reported by individual commentators and have not been independently verified, and that a mathematician called the moment the most significant in mathematical history.

Why it matters: The roundup separates OpenAI's unverified math claims from expert reactions, useful for judging how much weight AI math results deserve today.