Articles tagged "Open Weight"

**Google DeepMind's Gemma Team released the Gemma 4 Technical Report as [arXiv:2607.02770](https://arxiv.org/abs/2607.02770) on 2026-07-02**. The paper introduces **2.3B / 12B / 31B parameter models**, **both Dense and MoE variants**, **natively multimodal (text / image / audio)**, a **12B encoder-free unified design** (raw audio and image patches processed directly without separate encoders), a **built-in reasoning (thinking) mode**, **improved vision / audio encoders**, **architectural refinements for inference speed, memory efficiency, and long context**, and **competitive performance against larger open models on STEM, multimodal, and long-context benchmarks**. Over 300 authors contributed. Open weights allow commercial use, distributed via Hugging Face and Ollama. Sits alongside [Qwen 3.6-35B-A3B](../columns/qwen36-35b-a3b-uncensored-abliterated-2026-07) and the [Local LLM June 2026 update](../columns/local-llm-landscape-2026-june-update) as a new chapter at the open-weights frontier. **A milestone in Google's open-weights strategy**; the encoder-free unified design departs from Qwen / Llama / DeepSeek multimodality (separate vision encoder + projection). The **reasoning mode** mirrors the extended-thinking modes of Anthropic and OpenAI closed models — the open ecosystem catching up. Caveats: commercial-license fine print, potential systemic-risk classification (EU AI Act's 10^25 FLOPs threshold), and heavy Google Cloud Vertex AI integration bias.

**DeepReinforce released Ornith-1.0 on June 26, 2026** ([official](https://deep-reinforce.com/ornith_1_0.html) / [Hugging Face collection](https://huggingface.co/collections/deepreinforce-ai/ornith-10)). It is an **MIT-licensed open-weights family specialized for agentic coding**, **with no regional restrictions**. **Three sizes**: [Ornith-1.0-9B](https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B) (dense, ~19GB BF16) / [Ornith-1.0-35B](https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B) (MoE) / [Ornith-1.0-397B](https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B) (MoE, built on Qwen 3.5 + Gemma 4). **All sizes ship 262K context**, with **FP8 and GGUF quantizations released alongside**. **Benchmarks (vendor-reported, claimed SOTA at each open-weights size tier)**: | Benchmark | 9B | 35B | 397B | |---|---|---|---| | **SWE-Bench Verified** | **69.4%** | **75.6%** | **82.4%** | | **SWE-Bench Pro** | **42.9%** | **50.4%** | **62.2%** | | **SWE-Bench Multilingual** | — | — | **78.9%** | | **Terminal-Bench 2.1** | 43.1% | 64.2% | **77.5-78.2%** | | **NL2Repo** | 27.2% | 34.6% | **48.2%** | | **ClawEval** | — | — | **77.1%** | **Design thesis**: Reinforcement learning optimizes **both the solution rollouts and the scaffolding (the agent structure that drives them) itself** — a 'self-improving' agentic-coding design. It sits naturally next to the [Loop Engineering Maker-Checker](../columns/loop-engineering-ai-agent-paradigm-2026-06) paradigm. Reasoning is exposed via `<think>...</think>` blocks; function calling and tool use are first-class. **Distribution and ops**: vLLM ≥ 0.19.1 / SGLang ≥ 0.5.9 / Transformers ≥ 5.8.1 / Docker + llama.cpp / Ollama. OpenAI-compatible endpoints. The 9B fits on a single 80GB GPU; 35B and 397B want an **8×80GB GPU node (TP=8)**. Agent-framework compatibility: **OpenHands, Hermes Agent, and [OpenClaw](../services/openclaw-setup)** (Oflight's own service line — and ClawEval is in DeepReinforce's published benchmark set). **DeepReinforce lineage**: an RL-focused research organization that has previously shipped [CUDA-L1 (avg 3.12× GPU speedup)](https://github.com/deepreinforce-ai/CUDA-L1), [CUDA-L2 (HGEMM kernels beating cuBLAS)](https://github.com/deepreinforce-ai/CUDA-L2), and **IterX (MLSys 2026 NVIDIA Track)**. Ornith-1.0 applies the same RL playbook to LLM self-improvement. **Positioning**: alongside [Kimi K2.7-Code](../columns/kimi-k2-7-code-moonshot-ai-2026-06) (1T MoE / 32B active) and [GLM-5.2](../columns/local-llm-landscape-2026-june-update) (Intelligence Index v4.1 = 51, open-weights leader), **Ornith-1.0 is at the front of the June-2026 agentic-coding open-weights race**. Against Chinese-origin models (Kimi / GLM), its differentiator is **MIT license + no regional restrictions + a US-flag procurement story**. **Caveat**: benchmarks are DeepReinforce's own vendor-reported numbers. Independent third-party verification on public leaderboards has not yet appeared (as of June 26, 2026). The article closes with **three inquiry funnels for Ornith-1.0–era local-LLM evaluation, build, and ongoing maintenance**.

Two months after our [April 2026 local-LLM landscape column](../columns/local-llm-landscape-2026-april-comprehensive-comparison), here is the primary-source update on what has changed. **Three big shifts**: **(1) Open-weights have closed the gap with closed-source.** [GLM-5.2](https://simonwillison.net/2026/Jun/17/glm-52/) (Z.ai, MIT, June 16, 2026) tops the Intelligence Index v4.1 at **51** (MiniMax M3 44 / DeepSeek V4 Pro 44 / Kimi K2.6 43). [MiniMax M3](https://kilo.ai/open-source-models) ships **1M context + native multimodality + SWE-Bench Pro 59.0% + Terminal-Bench 2.1 66.0% + MCP Atlas 74.2%**. [NVIDIA Nemotron 3 Ultra](https://research.nvidia.com/labs/nemotron/Nemotron-3/) (revealed by Jensen Huang at Computex 2026) is a **550B-parameter** US-flag open-weight leader. [VibeThinker-3B](https://arxiv.org/pdf/2606.16140) (WeiboAI, MIT, Qwen2.5-Coder-3B fine-tune) reaches **frontier-reasoner parity at 3B**. **(2) Blackwell makes 30–70B models practical on consumer GPUs.** The RTX 5090 has **32GB GDDR7 and 1,792 GB/s bandwidth** (+77% vs 4090) with **native MXFP4 — GGUF Q4 runs with zero emulation overhead**, hitting **5,841 tok/s** on Qwen 2.5-Coder-7B at batch 8 (2.6× A100 80GB). The RTX PRO 6000 Blackwell reaches **~8,425 tok/s** on 30B; the B200 ships **192GB HBM3e at 8 TB/s** (4–5× H100). **(3) Japan's SI market is maturing.** **Intec** (TIS group) launched local-LLM deployment SI on January 29, 2026 — **minimum 1 month, from ¥5,000,000+ ex tax** — targeting manufacturing and finance. **Ricoh's 'RICOH On-Prem LLM Starter Kit'** won the **2025 Nikkei Excellent Product/Service Award grand prize** (Qwen2.5-VL-32B-Instruct base). PFN's [PLaMo 3.0 Prime](../columns/plamo-3-0-prime-pfn-japanese-llm-2026-06) was selected for the Japanese **Digital Agency 'Gennai'** common generative-AI platform — alongside the Mizuho / Lion Qwen on-domestic-infrastructure precedent. The column also covers concurrent moves on [Kimi K2.7-Code](../columns/kimi-k2-7-code-moonshot-ai-2026-06), [Sakana Fugu](../columns/sakana-fugu-orchestration-model-2026-06), [DiffusionGemma](../columns/diffusiongemma-google-text-diffusion-2026-06), and [Liquid AI LFM2.5-J](../columns/liquid-ai-lfm25-japanese-models-2026-06). Inference-engine selection (**AWQ + vLLM for GPU, GGUF + llama.cpp for CPU/edge, SGLang for agents, TensorRT-LLM for NVIDIA clusters**), quantization (BitNet 1.58-bit / MXFP4 / AWQ), regulation (**EU AI Act GPAI enforcement from August 2, 2026; systemic-risk threshold of 10^25 FLOPs**, US [Fable 5 export-control precedent](../columns/claude-fable-5-export-control-suspension-2026-06), Chinese-model cross-border data), typical GPU configurations by workload, and a three-step Oflight-recommended adoption path are all covered. The article closes with **three direct inquiry funnels** for local-LLM evaluation, build, and ongoing maintenance.