全部 AI 动态

🚀 Ling-3.0-flash-VL from @AntLingAGI is now live on Novita. Launching as a Day-0 partner ⚡ 🎁 Free for 14 days. Ling-3.0-flash-VL is a native multimodal MoE model built on Ling-3.0-flash, bringing visual information into understanding, reasoning, acting, and verification. 🔹 124B total parameters · 5.5B active parameters per token 🔹 Native image and video input 🔹 OpenAI-compatible serverless API 🔹 Function calling, reasoning, and Anthropic API support Built for multimodal reasoning, visual agent workflows, and long-context tasks.

🤗 Novita now supports Ling-3.0-flash-VL on @huggingface. 🎁 Free for 14 days. • 124B total parameters · 5.5B active parameters per token • Native image and video understanding • Built for multimodal reasoning and agentic workflows

inclusionAI's Ling-3.0-flash-VL brings vision to agent workflows, with BF16 and FP8 weights.📜 MIT. 🤖 https://modelscope.ai/models/inclusionAI/Ling-3.0-flash-VL 🤖 https://modelscope.ai/models/inclusionAI/Ling-3.0-flash-VL-fp8 🏆 Scores 42 on the Artificial Analysis Intelligence Index v4.1.1, up 4 points from Ling-3.0-flash. 👁️ Read charts and documents, count objects, and reason with visual evidence. Turn web and software interfaces into action sequences. 🎬 Understand images and videos. VideoRoPE tracks space and time for event localization, long-video Q&A, and video editing workflows. 🧠 124B total parameters, just 5.5B active per token. A hybrid KDA + Gated MLA backbone supports context up to 1M tokens. 📦 Choose BF16 or FP8 for lower weight memory. Deployment recipes cover SGLang and vLLM.

Meet Ling-3.0-flash-VL (124B/5.5B MoE) from @AntLingAGI, with native image + video understanding and a 1M context window. Day-0 support is live in SGLang! > Native multimodal: ViT encoder + MLP projector brings vision into the full loop of understanding, reasoning, acting, and verification > VideoRoPE: encodes space + time for event localization, long-video QA, and clip editing > 42-layer hybrid backbone: KDA + Gated MLA at 5:1 for efficient 1M-token long context Run Ling-3.0-flash-VL with SGLang!

More exciting news from @OpenAI: GPT-Image-2.5-Sunburst is #1 and Flare is #2 across the Text-to-Image Arena, the Image Edit Arena, and the Multi-Image Edit Arena! Both releases are significant improvements over GPT-Image-2 (medium) in all 3 Arenas. Sunburst delivers the highest quality generation, while Flare still improves upon GPT Image 2 at significantly faster speeds. The increase by points are (Sunburst | Flare): - Text-to-Image: +40 | +18 pts - Single-Image Edit: +59 | +30 pts - Multi-Image Edit: +81 | +47 pts Across all categories like 3D Imaging & Modeling, Art and more, Sunburst is #1 and Flare holds #2 in all three Arenas. Congrats to the team at @OpenAI!

This model represents a step-function improvement on many benchmarks, and its training is ongoing. Our internal model group arrived at the Navier–Stokes solution in 88 hours, using around 10,000 coordinating AI agents. Throughout the effort, we maintained the strict safeguards—including monitoring and isolation—that we apply to all our frontier evaluations.

🚀 Meet MiniCPM5-2B, a 2B-parameter language model bringing high intelligence density to the edge, now open source! It ranks #1 among open-source models under 4B parameters on the @ArtificialAnlys Intelligence Index, with a score of 23. It also scores 20 on the Agentic Index, bringing an early form of general-purpose agent capability to the edge. Across 34 benchmarks, MiniCPM5-2B achieves an average score of 53.9, covering coding, math, long-context understanding, tool use, and agentic tasks. And this release goes beyond the model itself. We’re opening up the data, training recipes, and RL stack behind MiniCPM5-2B. 🤗 Hugging Face: http://huggingface.co/openbmb/MiniCPM5-2B 💻 GitHub: http://github.com/OpenBMB/MiniCPM Modelscope: http://modelscope.cn/models/OpenBMB/MiniCPM5-2B Web: https://openbmb.cn/

OpenBMB’s MiniCPM5 brings 128K context, coding, reasoning, and agentic tool use to on-device models.📜 Apache 2.0. 🤖 https://modelscope.ai/collections/OpenBMB/MiniCPM5 🏆 Both 1B&2B models reach open-source SOTA in their size classes. MiniCPM5-2B also beats the larger models in the official comparison. 🛠️ Switch between fast answers and deep reasoning. Native tool calling powers local assistants and coding agents. 🧠 RL + OPD merges 16 specialist models, including five Agent experts, into MiniCPM5-2B. 📂 Open training data includes 500K Agent SFT samples and 80K+ RL samples for code, math, knowledge, and long-context reasoning. ⚡ Standard Llama architecture works with major inference engines. Official cookbooks and Agent Skills cover deployment and fine-tuning across popular frameworks.

🚀 Sol-H3: @MiniMax_AI H3 Video Generation Faster Than Playback 🤩 Five seconds of world. 1.653 seconds to infer. We’re releasing Sol-H3, our fastest end-to-end MiniMax-H3 inference stack yet. On one 8× NVIDIA B300 Blackwell system, it generates five seconds of 1344×768 video with stereo audio in 1.653 seconds. Across 1×, 4×, and 8× B300, Sol-H3 reaches up to a 15.54× speedup versus Base H3. Compared with 50-step Base H3 Dense on the same 8× B300 system, the four-step Sol-H3 profile delivers: • 5s: 18.250s → 1.653s (11.04×) • 10s: 50.660s → 3.732s (13.57×) • 15s: 99.513s → 6.612s (15.05×) Sol-H3 also scales across GPU counts: • 4× B300: 2.918s / 6.993s / 12.542s for 5s / 10s / 15s (12.11–15.54×) • 1× B300: 13.745s / 37.813s / 52.260s for 5s / 10s / 15s (9.45–14.29×) All figures are medians of three measured runs after one warmup at 1344×768 and 24 FPS with stereo audio. Base H3 uses 50 scheduler points (49 DiT forwards); Sol-H3 uses four DiT forwards, so this is a full-profile comparison—not an attention-only runtime change. Sol-H3 uses Dense attention on 1× B300 and SOL with INT8 QKV / FP8 output transport on 4× / 8×. Timing includes text encoding, DiT denoising, and video/audio VAE decoding; model loading, compilation warmup, and final MP4 encoding are excluded. Sol-H3 brings Sol-Engine × Sol-Attn into one full-stack runtime: • dynamic sparse attention with no retraining • fused norm, RoPE, MLP, and sparse-attention setup • fused INT8 QKV / FP8 output communication across 8 GPUs • parallel, batched VAE decoding • precomputed AdaLN caching Inside the stack: • sparse-attention setup: 1.206 → 0.285 ms (−76.4%) • VAE decode: 7.55 → 0.602 s • ~24 GB memory freed per GPU Any MiniMax-H3 few-step LoRA can plug into the same engine, and the code is deployment-friendly under Apache 2.0. For us, the bigger milestone is crossing from “fast generation” into “faster than playback.” That opens the path toward continuous 24 FPS generation and truly interactive video systems. We’re excited to partner with @reactorworld to release Sol-H3 and make it available as an API day-0. Try it now on Reactor: https://www.reactor.inc/sandbox?model=fast-h3 🔗 https://nvlabs.github.io/Sana/Sol-Engine/Sol-H3/ Amazing team effort—full credits in the blog. @shanasaimoe @lawrence_cjs @yitongli165665 @haopengl33 @HaochengXiUCB @songhan_mit

Congrats @OpenBMB on launching MiniCPM5-2B! SGLang has day-0 serving support, with recipes for @NVIDIAAI RTX PRO 6000, RTX 5090, DGX Spark, and Hopper cards. We are seeing over 250 tok/s/user decode at bs=1 on coding tasks with DSpark enabled on a 5090. It's an ultra-efficient 2B with incredible intelligence density. As of today it leads Artificial Analysis's Intelligence Index among open models under 4B, and scores above models more than 4x its size. This is another step toward general-purpose models running 24x7 on device, even on phones. Run MiniCPM5-2B with SGLang 👇