Open Source

NVIDIA has open-sourced OSMO, the Kubernetes-native workflow orchestrator it uses internally for Project GR00T, Isaac Lab, and Isaac Sim. OSMO lets robotics teams define training, simulation, and hardware-in-the-loop tasks in a single YAML file and routes each one to the right compute tier, from GB200 clusters to Jetson AGX Thor devices, without infrastructure code. Apache-2.0 licensed, latest release 6.3.1.

Pizza Bot is an open source, self-hosted inbox for AI agents built on DeepAgents and LangGraph. It combines persistent task state, MCP integrations, configurable approvals, and scheduled workflows across multiple model providers.

Long-horizon agents have turned LLM serving into an input-heavy workload. Repeated prefills and million-token contexts leave KV caches that strain HBM, SSD capacity, and...

Robot datasets have grown far slower than the models trained on them, mostly because collection stays locked to lab hardware. AXIS moves demonstration collection into a web browser and pushes everything expensive to backend GPUs. The result is 207 tasks and 50,129 verified Franka trajectories, and continual pretraining that lifts π0.5 from 83.9 to 88.8 on LIBERO-Plus while a volume-matched RoboCasa365 control reaches only 57.5.

Most open model launches release one checkpoint and a benchmark table. The Institute of Foundation Models (IFM) released something wider last week. IFM is...

We look at NeoMME, a family of 260M and 800M bidirectional encoders from H Company. Unlike ColPali-style retrievers, it processes multilingual text tokens and raw 32×32 image patches in a single Transformer, with no pretrained vision tower and no causal decoder. We cover the masked discrete-diffusion pretraining objective, the dual dense and late-interaction retrieval heads, and the ViDoRe v3 results where the 260M model reaches 0.523 nDCG@10. We also break down the 255× index compression, the 51.3 pages per second indexing throughput on one L40S, and the text-retrieval gaps the authors acknowledge.

Training and benchmarking a computer-use agent needs four things — agents, environments, traces, and a framework to evaluate and train them — and all four ship in incompatible formats today. CUA-Lite, from a UC Berkeley led team, puts them behind one action space and one data schema, and replaces OSWorld's per-task virtual machine with a plain Docker container at 0.9 GB instead of 4.1 GB.

Perplexity has open sourced Lily, the local inference engine behind Hybrid Compute in Perplexity Computer. Built in Rust with custom Metal kernels for one model on one chip family, it averages 1.23x MLX-LM's prefill throughput and 1.35x its decode throughput on a 40-core, 128 GB M5 Max.

We look at zg (zvec-grep), the local-first search layer that Qwen Developers just open-sourced under Apache 2.0. We explain how it puts ripgrep, BM25, and vector search behind a single interface, so an agent can move from a plain-language description to an exact line span without switching tools. We cover the deliberately small MCP surface, the on-device embedding catalog, and the authorization gate that stands between local content and remote models.

NVIDIA has released Switchyard, an Apache-2.0 Rust proxy and library for LLM traffic. It decodes requests into provider-neutral types, routes them with passthrough, random, LLM-classifier, or stage-router algorithms, and translates responses back into the client's format, so Claude Code or Codex CLI can run against vLLM, NIM, or Ollama unchanged. The project is pre-alpha and not for production.

How do you benchmark a web search API when the thing being tested can read the answer key? A search agent has a fetch...

Google Research has released TimesFM-3, a 330 million parameter time series foundation model that forecasts multiple related series in a single forward pass. Unlike every TimesFM checkpoint through 2.5, it is pretrained natively for multivariate forecasting, accepting multiple targets, past covariates, and past-future covariates with no task-specific fine-tuning. It takes the top average rank among pretrained foundation models on GIFT-Eval, fev-bench, and the TIME leaderboard. The weights, however, ship under a non-commercial, non-production license.

The OpenClaw Foundation has released v2026.8.1, which the project calls OpenClaw 2.0: 933 contributors, 569 first-timers, and more than 16,000 pull requests, roughly half of every PR ever merged into the repo. Setup now reuses existing subscriptions, API keys and local models. The rebuilt Control UI cut test-harness startup from about 1.6 seconds to 575 ms. Shared cloud sessions add real multiplayer, though the docs are explicit that they are not a security boundary.

Google Cloud AI Research, with Washington University in St. Louis and UNC Chapel Hill, has released EnvHarness, an Apache-2.0 layer that turns a static agent benchmark into one that adapts to the policy training on it. It wraps a frozen environment through the standard reset()/step() interface, so tasks and human-built verifiers stay untouched — and an LLM designer, EnvRigger, writes those wrappers automatically against flaws diagnosed in the agent's own rollouts. Across five benchmarks, mined skills gain up to 9.0 points on held-out tasks with 9.8% fewer execution steps.

Vercel has open-sourced vgpu, the WebGPU library it built to ship the shaders on vercel.com. It treats .wgsl files as importable TypeScript modules, runs the same shader in the browser, in headless Node.js via Dawn, and in a deterministic CI mock, and ships a fullscreen effect in 25 KB gzipped.

Model cards report quality under server-class, full-precision conditions. Those numbers rarely predict how the same model behaves on a phone. This week, Liquid AI...

Three ~300M drafters bring speculative decoding to LFM2.5, delivering up to 3.18x faster decoding with identical greedy output.

NVIDIA has released TensorRT Model Connect (TRTMC) in public preview, an Apache-2.0 project that takes a supported Hugging Face or local checkpoint to end-to-end TensorRT inference in two commands, with no intermediate ONNX export. The build emits a versioned .bundle artifact that runs through native C++ task APIs, so inference executes without PyTorch in the runtime path. NVIDIA's July 29, 2026 GB300 snapshot covers 105 release profiles across 76 model families.

Google has open-sourced SAM (Sovereign Agent Mesh) under Apache-2.0 — and it has nothing to do with Segment Anything. SAM is a zero-config, zero-trust P2P overlay that lets autonomous agents discover and call each other's MCP tools across cloud, on-prem, laptop and edge environments, without exposing a single internal endpoint to the public internet. Identity flows from OIDC into Biscuit capability tokens, so nodes authorize every request offline under a strict default-deny model.

Nous Research has shipped Bot Mode for Hermes Agent, its MIT-licensed open source agent. Bot Mode replaces the single-agent session list with a roster of named bots. Each bot is a real Hermes profile, with its own chat, memory, skills, and pinned model. It is now bundled and default-on in Hermes Desktop.

MiniMax released MiniMax-Music3, an open-weights text-to-music model. Given lyrics with section tags and a structured caption, it generates a complete song of up to five minutes in a single pass, as 32 kHz, 16-bit stereo WAV. Here is the architecture, the three serving paths, and the license conditions that matter before you ship.

DeepSeek Harness v0.1 is an MIT-licensed agent harness where every capability is a Cordis plugin. Four runtime modes, append-only session logs, and provider-agnostic model routing.

Cactus Compute released Needle 2, an open 45M-parameter model for tool calling, device use, and structured extraction. The full model is a single 14MB binary that runs a session in about 28MB of RAM. It leads both Seal-Tools splits while targeting hardware with no GPU and no NPU.

Liquid AI released LFM2.5-VL-3B, a 3.1B-parameter vision-language model built for on-device deployment. It averages 80.7 on ScreenSpot-v2 and lifts RefCOCO grounding from 57.1 to 87.9. Function calling is new to the VL line, with ToolSandbox moving from 26.4 to 59.5. The model fits in roughly 3 GB and decodes 228 tokens/s on an Apple M5 Max.

NVIDIA's open 30B MoE targets the agent execution layer, with Switchyard routing each step to the cheapest capable model.

LTX-2.5 brings frontier video generation to local NVIDIA hardware: 6.8-second clips, native multishot, day-one ComfyUI, open weights.

Meta's Muse Glimmer is a 30B open-weights agentic model under Apache 2.0. It fits 24 GB VRAM and decodes 3.1x faster with DFlash speculation.

NVIDIA releases NemotronLabs VoiceChat 11B, an open full-duplex speech-to-speech model with 448 ms latency and live tool calling.

Long agent runs accumulate state that no transcript records — edited files, a live dev server, installed packages, a warm prompt cache. When an agent misreads a traceback at step 10 and rewrites a correct file, patching forward burns tokens and restarting re-pays every call. Researchers at Northeastern University and Stanford University released Shepherd, an MIT-licensed Python runtime substrate that records every agent-environment interaction as a typed event in a Git-like execution trace. Each commit covers the agent process and filesystem together, copy-on-write, so a rewind restores live state instead of just files. The paper reports 5× faster forks than Docker, over 95% prompt-cache reuse on replay, and a live supervisor raising CooperBench pair-coding pass rates from 28.8% to 54.7%.

Mistral AI has released Shieldstral 1.0 3B, an open-weights, policy-adaptive multimodal safety classifier that frames content moderation as a single yes/no question instead of a fixed harm taxonomy. Operators supply the policy as a plain-language query at inference time and get back a calibrated safety score from one forward pass — no retraining required to re-target the model. Built on Ministral-3-3B-Base-2512 with a Pixtral vision encoder and trained on roughly 54.1M samples, it reports 84.9% average F1 on text safety (matching GPT-OSS-Safeguard-20B), 83.8% on multimodal safety, and 91.3% on Mistral's adaptability benchmark — while fitting in 16GB of VRAM under an Apache 2.0 license.

Tencent Cloud has open-sourced TencentDB Agent Memory v2.0, a team-level memory hub that turns conversations, documents and code into four governed, reusable assets — Chat Memory, Skill, LLM-Wiki and Code-Graph. It is MIT-licensed, self-hosted via Docker, and integrates with Claude Code, OpenClaw, Hermes and CodeBuddy. The differentiator is not retrieval but governance: ACL-based visibility decides which agent gets which asset, and which version is valid.

NVIDIA Labs has open-sourced NOOA (NVIDIA Object-Oriented Agents), a model-agnostic Python framework for building AI agents. Agent development today is split across prompt templates,...

Microsoft has open sourced code-testing-generator, a polyglot unit-test agent shipping in the MIT-licensed dotnet/skills repository. It reads a repository before writing anything — detecting the language, test framework, existing conventions, and the real build and test commands — then plans, writes, runs and validates the tests it produces. On Microsoft's internal 152-task benchmark it completed 140 tasks against 120 for stock GitHub Copilot on the same model, with the gain concentrated almost entirely in vague prompts and diff-targeted requests.

Prime Intellect has open-sourced Prime Agent, a coding and research harness built on two abstractions: the Recursive Language Model, which turns sub-agent calls into functions inside a persistent IPython kernel, and the Continual Harness, which lets the agent edit its own prompts, skills, memory, and sub-agent specs mid-run. With Opus 5 it reports 95.5% RHAE Best@1 on ARC-AGI-3, above the reported human expert baseline of 95.4%.

NVIDIA released Alpamayo 2 Super, a 34B vision-language-action model for autonomous driving, under OpenMDW-1.1 — a permissive license covering fine-tuning, derivatives and commercial redistribution. It pairs a 32B Cosmos 3 Super Reasoner backbone with a 2.3B diffusion action decoder, scores 79.2 on LingoQA, and emits trajectories, Chain-of-Causation traces, meta-actions, auto-labels and grounded VQA from a single pass.

CopilotKit has published the Channels SDK, an MIT licensed library that runs an existing AG-UI agent inside Slack and Microsoft Teams. Version 0.5.0 ships five platform adapters and a documented runtime contract. This breakdown covers the verified deployment paths, the baseline requirements, and the one dependency that is easy to miss

Cursor Research has open-sourced Mixture-of-Kittens (MoK), the MoE training megakernel behind its Composer models. MoK fuses all mixture-of-experts communication and computation into a single deterministic kernel, and runs up to 2.37x faster than the strongest public baseline on GB300 NVL72 racks. It requires Blackwell SM100 or SM103 GPUs, which puts it out of reach for anyone without NVL72 capacity.

Reflex has released XY, an Apache-2.0 Python charting library that moves rendering work into a native Rust core and a WebGL2 client. It holds roughly 0.08 seconds render time from 10,000 to 100 million points, exports a 10-million-point interactive scatter at 258 KiB, and keeps exact f64 columns in Python so hover, selection, and zoom drilldown still return original rows. The library is early alpha at version 0.0.1.

Y Combinator has open-sourced QM, the multiplayer agent harness it uses internally across accounting, legal, events, and engineering. Released July 31, 2026 under an MIT license, QM gives each employee an isolated workspace and each Slack room its own scoped memory, files, keychain view, permissions, crons, web apps, and durable sandbox. Pi, OpenCode, Codex, and Claude Code all drive the same headless core, so deployments avoid vendor lock-in.

Genspark has open sourced GenOffice under the Apache License 2.0. It is an AI-native office suite for macOS and Windows, covering Docs, Sheets, Slides and PDF as five Electron apps over one shared engine layer. The notable engineering claim is a byte-preserving round trip: only edited paragraphs are regenerated as OOXML and spliced back into the original file, so untouched blocks keep their original bytes and layout survives in Word. Sheets pairs the open-source Univer core with an in-house Rust xlsx sidecar. AI calls route through a signed-in Genspark account and consume credits. The repository labels this an Alpha.

AMD released Instella-MoE-16B-A3B, a fully open Mixture-of-Experts language model trained from scratch on Instinct MI300X and MI325X GPUs. It holds 16B total parameters but activates only 2.8B per token, using Gated MLA and FarSkip-Collective. AMD published weights from every training stage, plus data mixtures, configs, and inference code.

Supabase has open sourced supabase/evals, an Apache-2.0 benchmark and framework that runs coding agents including Claude Code, Codex and OpenCode against real Supabase tasks — building schemas, debugging Edge Functions, fixing RLS policies — inside containerized stacks, then scores them with deterministic checks and LLM-as-a-judge.

JetBrains Research has open-sourced KotlinLLM under the Apache License 2.0. The IntelliJ IDEA plugin prototype adds Smart macros, asLlm and mockLlm, whose bodies are generated Kotlin source rather than live model calls. The plugin captures runtime values through JDI, asks an LLM agent for a narrow code update, compiles it, and redefines the loaded class. Covered scenarios then run as plain Kotlin with no further inference call. On an adapted Spring Petclinic project, 24 of 24 scenarios completed with a 100% hot-reload success rate and roughly 1% runtime overhead.

Tencent has released AngelSpec, an open-source torch-native framework for training speculative-decoding draft models across six architectures. It introduces DFly, a block-diffusion drafter with hybrid target conditioning and a hidden-correction autoregressive head, and integrates D-cut for runtime-adaptive verification budgeting. On HY3-295B-A21B with TP=8, DFly-8 delivers a 1.98–2.40× speedup over autoregressive decoding across concurrency 4 to 64.

Marktechpost AI has released Token Saver, an open-source MCP extension for Claude Desktop that uses local Hybrid RAG to slash PDF token consumption by up to 99% while ensuring absolute document privacy.

Moonshot AI has open-sourced MoonEP, an Expert Parallelism (EP) communication library for distributed Mixture-of-Experts (MoE) workloads. The team announced the release as a library...

Liquid AI released two open-weight bidirectional encoders, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M. Both carry an 8,192-token context and are built on the LFM2 hybrid backbone. The 350M ranks fourth of 14 models on a 17-task GLUE, SuperGLUE, and multilingual suite, behind only larger models. The 230M clears one 8K-token forward pass on CPU in about 28 seconds.

Moonshot AI's Kimi team and kvcache-ai open-sourced AgentENV (AENV) under MIT, as part of Kimi K3 Open Day. It runs agent sandboxes as Firecracker microVMs with millisecond snapshot, resume, and 16-way fork, behind an E2B-compatible API.

Perplexity has released pplx, an official command line client for its Search API. The tool exposes two commands — pplx search web and pplx content fetch — and returns exactly one JSON object on stdout. It ships as a checksum-verified single binary for macOS arm64 and Linux, alongside an Agent Skill for Claude Code, Codex CLI and any harness that can read a URL.

Datalab rewrote Marker as a three-mode pipeline. Version 2 hits 76.0 on olmOCR-bench and sustains 2.9 pages per second on one B200 — over 5× MinerU's pipeline backend, while beating Docling on both accuracy and speed. Here's how it compares against MinerU, Docling and LiteParse, and which one fits your use case.

Andrew Ng has released OpenWorker, an MIT-licensed desktop AI agent that returns finished deliverables instead of chat replies. It runs a local Python agent server under a Tauri shell, supports 30 curated tool-calling models plus fully local Ollama, and gates every write, shell command and off-machine action behind a typed risk engine.

Gigatoken is an MIT-licensed Rust BPE tokenizer that encodes text at gigabytes per second. On a 144-core AMD EPYC 9565, it runs GPT-2 tokenization at 24.53 GB/s. That is 989x HuggingFace tokenizers and 681x tiktoken, against baselines already written in multithreaded Rust. The gains come from a hand-written SWAR pretokenizer and pretoken caching, not a faster BPE merge loop.

Cisco Foundation AI has released Antares, a family of small language models trained to pinpoint where known vulnerabilities live inside a codebase. Antares-1B reaches 0.209 File F1 on the new Vulnerability Localization Benchmark, above GLM-5.2 at 753B parameters and Gemini 3 Pro. The untrained Granite 4.0 checkpoints score near zero under the same protocol, so post-training supplies almost all of the capability. A full 500-task sweep runs in roughly 13 minutes on a single H100 for under a dollar, against $141 for GPT-5.5.

Poolside has released Laguna S 2.1, a 118B open-weight Mixture-of-Experts coding model with 8B active parameters per token and a 1M-token context. It matches or beats models several times its size on agentic coding benchmarks, ships under OpenMDW-1.1, and runs on a single NVIDIA DGX Spark.

Meta has open-sourced Astryx, the React and StyleX design system it ran internally for eight years across 13,000+ apps. It ships 150+ accessible components, seven themes, dark mode, templates, and an agent-ready CLI under MIT — with React 19+ required.

NVIDIA DeepStream 9.1 introduces 13 agentic skills that let coding agents like Claude Code and Codex build multi-camera video analytics pipelines from natural-language prompts. Multi-View 3D Tracking (MV3DT) fuses per-camera detections into one shared 3D world with a globally consistent object ID, while AutoMagicCalib (AMC) removes manual camera calibration. The release also adds JetPack 7.2 support and a unified open-source GitHub monorepo.

Google Cloud's generative-ai repository ships the Always-On Memory Agent, a reference implementation that treats memory as a running process. Built on Google ADK and Gemini 3.1 Flash-Lite, it uses no vector database and no embeddings. Instead, an orchestrator routes to Ingest, Consolidate, and Query sub-agents that read, connect, and write structured memory into SQLite 24/7.

Zyphra released ZUNA1.1 on July 16, 2026, under the Apache 2.0 license. The 380M masked diffusion autoencoder reconstructs, denoises, and upsamples scalp-EEG across arbitrary channel layouts. It accepts variable-length inputs from 0.5 to 30 seconds, against ZUNA1's fixed five seconds. Reported NMSE holds or improves while the input range widens.

NVIDIA released Nemotron 3 Embed on July 15 and 16, 2026. The collection has three open checkpoints: Nemotron-3-Embed-8B-BF16, Nemotron-3-Embed-1B-BF16, and Nemotron-3-Embed-1B-NVFP4. The 8B ranks #1 on RTEB at 78.46 average NDCG@10. The 1B came from ModelOpt NAS pruning plus COS+MSE distillation from the 8B teacher. NVFP4 retains 99%+ of BF16 retrieval accuracy at up to 2x Blackwell throughput. All three run 32,768-token inputs under OpenMDW-1.1.

SpaceXAI published the Grok Build source on July 15, 2026. The Apache 2.0 Rust tree covers the agent loop, tool dispatch, the TUI, and the extension system. Grok 4.5 stays closed, and external contributions are not accepted.

Thinking Machines Lab released Inkling on July 15, 2026, its first model trained from scratch. The full weights ship under Apache 2.0. It is a 975B-parameter Mixture-of-Experts transformer with 41B active parameters, a 1M-token context window, and native text, image, and audio input. The lab states plainly that Inkling is not the strongest model available, open or closed. It is positioned instead as a customization base, with controllable thinking effort as the practical differentiator.

Soofi S 30B-A3B is an open Mamba-Transformer MoE model activating 3.2B of 31.6B parameters for German and English

Developer Hayden Bleasel has released Blume, an open-source, MIT-licensed documentation framework. It reads a folder of Markdown or MDX and generates a hidden Astro project, shipping static, AI-ready docs with local search, 30+ MDX components, llms.txt, and a built-in MCP server.

Robbyant, Ant Group's embodied-intelligence unit, has released LingBot-World-Infinity (LingBot-World 2.0). It is a 14B causal video generation model that behaves as an interactive world simulator. The core idea is the Mixture of Bidirectional and Autoregressive (MoBA) attention mask, paired with distribution matching distillation applied over long self-rollout trajectories. Together they target long-horizon drift, the failure mode that smears textures and warps geometry in most interactive world models. A Director-Pilot agentic harness wraps the generator, where a VLM proposes events and the Diffusion Transformer renders them. The report shows a single 60-minute uninterrupted session covering 20 scenarios. But the release is thinner than the paper: one checkpoint, a 480P reference script, no deployment code, no quantitative benchmarks, and a non-commercial CC BY-NC-SA 4.0 license.

Ant Group's Robbyant has released LingBot-VLA 2.0, an Apache-2.0 vision-language-action model for cross-embodiment robot manipulation. The 6B checkpoint is pretrained on roughly 60,000 hours of data, spanning 50,000 hours of robot trajectories across 20 robot configurations and 10,000 hours of egocentric human video. It maps every embodiment into a single 55-dimensional canonical action space, covering arms, dexterous hands, waists, heads, and mobile bases. A token-level, auxiliary-loss-free Mixture-of-Experts action expert scales capacity without adding a load-balancing loss. Dual-query distillation from LingBot-Depth and DINO-Video adds geometric and temporal supervision for future-aware control. On the GM-100 generalist benchmark it outperforms π0.5 and LingBot-VLA-1.0 on both evaluated platforms.

Ant Group's Robbyant open-sourced LingBot-Vision, a self-supervised ViT family for dense spatial perception. Masked boundary modeling makes image boundaries a native training signal. The 1B backbone matches or surpasses larger models, and initializes LingBot-Depth 2.0.

Liquid AI released Antidoom, an open-source method that targets doom loops in reasoning models. A doom loop repeats a span until the context window is exhausted. Antidoom finds the token that starts the loop and retrains only that position using Final Token Preference Optimization (FTPO). On LFM2.5-2.6B, doom-loop rates fell from 10.2% to 1.4%; on Qwen3.5-4B, from 22.9% to 1%. Generation, detection, and the FTPO trainer are open source.

Synthetic Sciences has released OpenScience, an Apache-2.0 AI workbench for scientific research. It works with any frontier or open-weight model, using your own API keys. It runs the full loop across machine learning, biology, physics, and chemistry. It ships 250+ editable skills and queryable scientific databases, and it runs on your own infrastructure.

LlamaIndex’s legal-kb is a public reference app that gives agents filesystem-style access to a document knowledge base on Index v2. It exposes retrieve (hybrid semantic search), find, read, and grep as tools, with automatic per-file versioning and visual citations. The stack is TanStack Start, AI SDK 6 (ToolLoopAgent), Prisma, and WorkOS.

Most enterprise data still sits inside PDFs, scans, and slide decks. Large language models and agents cannot use that data until it becomes structured...

WebBrain is a free, MIT-licensed AI browser agent for Chrome and Firefox. It reads pages, extracts data, and automates multi-step tasks through Ask and Act modes. Run it on local models like llama.cpp or Ollama for privacy, or connect any cloud API.

Interfaze open-sourced diffusion-gemma-asr-small, a multilingual ASR model that transcribes via diffusion, not autoregression. It adds audio to Google's frozen DiffusionGemma using a ~42M-parameter adapter. One adapter covers six languages, with transcription cost set by denoising steps, not transcript length.

The Google Health API now has an open-source CLI. ghealth is a single Go binary that exposes 40 data types as agent-ready JSON. It is a community project, not an official Google release. Here's how it works, and what to check before granting OAuth access.

NVIDIA has released Nemotron-Labs-TwoTower, a diffusion language model built on a pretrained autoregressive backbone. It ships as open weights under the NVIDIA Nemotron Open...

Google Research has released TabFM, a foundation model for tabular data. It performs zero-shot classification and regression through in-context learning. Predictions come from a single forward pass, with no per-dataset training, hyperparameter tuning, or feature engineering.

EverMind has open-sourced EverOS, a local-first memory runtime that stores AI agent memory as plain Markdown indexed by SQLite and LanceDB. It combines hybrid BM25 + vector retrieval, multimodal ingestion, and self-evolving Skills under an Apache 2.0 license. Here's what it is, how the architecture works, where the benchmarks stand, and where it still falls short — plus a runnable code walkthrough and an interactive demo.

Liquid AI released LFM2.5-230M, its smallest model yet. The 230M-parameter, open-weight model runs on-device at 213 tok/s on a Galaxy S25 Ultra and 42 on a Raspberry Pi 5. Built on the LFM2 architecture, it targets tool use and data extraction, beating larger models like Qwen3.5-0.8B and Gemma 3 1B on instruction following.

DeepSeek open-sourced DSpark, a speculative decoding framework that attaches a draft module to existing DeepSeek-V4 weights. It pairs a parallel draft backbone with a lightweight Markov head to cut suffix decay, then adds confidence-scheduled verification that tailors how many tokens get checked to real-time GPU load. Offline, accepted length rises 16–31% over DFlash and Eagle3; in production it speeds per-user generation 57–85% over the MTP-1 baseline, losslessly. The training repo, DeepSpec, ships under MIT.

Apple released container 1.0, an open-source Swift tool running Linux containers as lightweight virtual machines on Apple silicon.

DeepReinforce released Ornith-1.0, an open-source coding model family built on Gemma 4 and Qwen 3.5. Instead of a fixed harness, the model learns its own scaffold during reinforcement learning. The 397B flagship reports 82.4 on SWE-Bench Verified, with all weights under the MIT license.

Baidu open-sourced Unlimited OCR, a 3B-parameter MoE model that parses dozens of document pages in a single forward pass. Its Reference Sliding Window Attention (R-SWA) holds the KV cache constant, so memory and latency stay flat as output grows. It scores 93.23 on OmniDocBench v1.5, beating the DeepSeek OCR baseline by 6.22 points, under an MIT license.

UC San Diego's DFlash replaces autoregressive drafting with a lightweight block diffusion model for speculative decoding. It drafts whole token blocks in a single forward pass and conditions on target hidden features through KV injection. The paper reports up to 6.08x lossless speedup on Qwen3-8B, while NVIDIA reports up to 15x throughput on Blackwell at fixed interactivity. DFlash ships 20 checkpoints and supports SGLang, vLLM, and TensorRT-LLM.

Datalab released lift, a 9B open-weights vision model that turns PDFs and images into schema-matching JSON. It uses schema-constrained decoding for valid structure and trained abstention to return null instead of hallucinating absent fields, scoring 90.2% field accuracy on a 225-document benchmark.

Prime Intellect has released prime-rl 0.6.0, an open framework for asynchronous reinforcement learning on trillion-parameter Mixture-of-Experts models. It trained GLM-5 on SWE tasks at up to 131k sequence length, with sub-5-minute step times and 256 rollouts, on 28 H200 nodes. This breakdown covers the inference and training optimizations behind those numbers — FP8 inference, Wide Expert Parallelism, prefill/decode disaggregation, router replay, and 3-D parallelism (FSDP, EP, CP).

The HIP kernel uses one-instruction asm wrappers and an eight-wave pipeline to outperform AMD's AITER v3 on MI300X.

Cisco Foundation AI has open-sourced FAPO (Fully Automated Prompt Optimization), a Claude Code-driven system that autonomously optimizes multi-step LLM pipelines from baseline prompts to target accuracy. FAPO evaluates a chain, attributes failures at the step level, proposes variants across prompt, parameter, and chain-structure levels, and validates each through an independent reviewer. In Cisco's evaluation, it beat GEPA on 15 of 18 model-benchmark comparisons. Here's how the optimization loop works and how to run it.

Nous Research has added a Blank Slate setup mode to its open-source Hermes Agent. It starts an agent with everything off except provider, model, File Operations, and Terminal. You opt in to the rest.

In this article we look at YaFF, Yandex's open-source zero-copy wire format for the Protobuf ecosystem. We keep the .proto file as the single source of truth, changing only how data sits in memory. We walk through its four layouts — Fixed, Flat, Sparse, and Dynamic — and the benchmark where Flat Layout reads within 1.2× of a raw C++ struct. We also cover where it fits, including the advertising recommendation system reporting 10–20% CPU savings at production scale.

SpatialClaw is a training-free agent that writes Python in a persistent kernel, composing perception tools for 3D spatial reasoning

VibeThinker-3B, a 3B MIT-licensed reasoning model matching DeepSeek V3.2 and Kimi K2.5 on verifiable benchmarks.

Vercel has open-sourced eve, an Apache-2.0 agent framework now in public preview. An agent is a directory of files, with durable execution, sandboxes, approvals, connections, channels, and evals built in. Scaffold with npx eve@latest init and deploy unchanged via vercel deploy.

Flash-KMeans is an open-source, IO-aware implementation of standard Lloyd's k-means in Triton GPU kernels. It does not change the math or approximate. FlashAssign removes distance-matrix materialization; Sort-Inverse Update eliminates atomic contention. On an NVIDIA H200, it reports 17.9× end-to-end, 33× over cuML, and over 200× over FAISS.

Databricks has open-sourced Omnigent, a meta-harness that sits above coding agents like Claude Code, Codex, and Pi. It adds composition, contextual policies, and live session sharing under one interface, on terminal, web, desktop, and mobile. The Apache 2.0 project is in alpha.

Moonshot AI has open-sourced Kimi K2.7-Code under a Modified MIT license. It is a coding-focused, agentic model built on Kimi K2.6, with a 256K context window and roughly 30% lower reasoning-token usage. Moonshot reports gains over K2.6 on six benchmarks, including +21.8% on Kimi Code Bench v2. The model is available via the Kimi API and Kimi Code.

We look at Gemini-SQL2, the text-to-SQL capability Google Research announced on June 12, 2026. Powered by Gemini 3.1 Pro, it posted 80.04% execution accuracy on the BIRD single-model leaderboard. We explain what the score measures, how the leaderboard stacks up, and what Google has not yet disclosed. We also cover use cases and a schema-grounded implementation pattern.

Zyphra has released Zamba2-VL, a family of open vision-language models at 1.2B, 2.7B, and 7B parameters. The models use a hybrid Mamba2 state-space and Transformer backbone, shipping under Apache 2.0. They stay competitive with comparable Transformer VLMs while cutting time-to-first-token by about an order of magnitude.

Cohere's first developer coding model is a 30B mixture-of-experts running on a single H100 with 256K context length.

DiffusionGemma is Google DeepMind's experimental 26B open model using text diffusion for up to 4x faster generation on GPUs.

UIUC and Chroma's Harness-1 is a 20B retrieval subagent trained with reinforcement learning inside a stateful search harness. The harness maintains the bookkeeping — candidate pool, importance-tagged curated set, evidence graph, verification records — while the policy decides what to search, curate, verify, and when to stop. It reaches 0.730 average curated recall across eight benchmarks, beating the next open subagent by 11.4 points and trailing only Opus-4.6. Weights and harness code are public.

Google released the Colab CLI, letting developers and AI agents run local code on remote Colab GPU and TPU runtime