🆕 A new open-source 1.6 TRILLION-parameter MoE that scores 0.1 points behind Claude Opus 5 on an agent benchmark has been released. @NexEcosystem released Nex-N2.5-Max. It beats DeepSeek V4 Pro 0813 on all but one comparable benchmark. This thing is enormous! Stats ... 🧠 1.6 TRILLION parameters 🔀 Mixture-of-Experts 🤖 Built for agents + long-horizon workflows 💻 Coding + tool use 📚 262K context 📝 Text-only ⚖️ Apache 2.0 📦 Weights are already on Hugging Face Nex's reported results 👇 AutomationBench 🟣 Claude Opus 5 → 50.3 🔥 Nex-N2.5-Max → 50.2 🔵 GPT-5.6 Sol → 45.8 🟢 Qwen3.8-Max → 39.8 And on BrowseComp: 🔥 Nex-N2.5-Max → 92.6 🟣 Claude Opus 5 → 90.8 Terminal-Bench 2.1 → 86.1 Toolathlon Verified → 74.7 😁 You're probably NOT running this one under your desk. The official model repository is about 1.65TB, and Nex's recommended deployment uses this monster hardware .... 🎮 16× H200 🖥️ 2 nodes ⚡ TP16 + expert parallelism But here's why I'm posting the monster first. It has a 35B multimodal little brother. And THAT one may actually belong in your home AI lab. 👀🔥

Announcing Fugu-Ultra v1.1 🐡 We’ve been thrilled by the reception to the Fugu model family. Thanks to everyone who tried it, shared feedback, and trusted Fugu with real work. Today, we’re releasing Fugu-Ultra v1.1 → Upgraded to incorporate the latest frontier models, resulting in stronger performance across every benchmark shown, including gains of up to 7.9 points over v1.0, with particularly strong results on ProgramBench and Terminal Bench 2.1. Fugu-Ultra v1.1 is more capable across coding, agentic tasks, and advanced reasoning, and available at the same price as Fugu-Ultra v1.0 The frontier keeps moving, and Fugu keeps getting better.

Announcing Artificial Analysis Intelligence Index v4.3, upgrading Terminal-Bench to 4.0 and adding AutomationBench-AA, an agentic workflow automation benchmark with a private test set. This is a continuation of our rollout of Intelligence Index v5 Changelog (Index v4.2 → Index v4.3): ➤ Terminal-Bench: 2.1 → 4.0, completing our upgrade to the latest version of Terminal-Bench ➤ Replacing 𝜏³-Banking with AutomationBench-AA, our implementation of Zapier's business workflow automation benchmark We are continuing to prioritize keeping Intelligence Index as useful as possible by bringing forward a subset of the changes we had planned for Index v5. Each change in v4.2 and v4.3 stands on its own merits and brings the Index closer to real-world problem solving, adds more private test sets to prevent gaming, and reduces saturation Intelligence Index v4.3 raises the difficulty of agentic coding tasks and broadens the types of agentic workflows tested. Because we use a held-out test set for AutomationBench-AA, in collaboration with @zapier, the weight assigned to evaluations with private tasks or answers increases from 40% to 45%. Category weights are unchanged from v4.2: Agents 30%, Coding 20%, General 30%, Scientific Reasoning 20% Detailed changes: ➤ Upgraded Terminal-Bench 2.1 to 4.0: 66 multi-step tasks testing agents on tasks run in agent sandboxes driven via the terminal, including tasks involving software engineering, machine learning, science, and operations. The 4.0 update recalibrates compute and time allowances, and improves task instructions and verification. We have changed from the Terminus 2 harness to mini-SWE-agent, a minimal, model-agnostic harness. We will also be updating our Coding Agent Index, where we test model and harness pairs, to include Terminal-Bench 4.0 soon ➤ Replaced 𝜏³-Banking with AutomationBench-AA: Our implementation of Zapier’s AutomationBench tests agents on 657 business workflows across simulated applications such as Gmail, Slack, Salesforce, and Jira. Agents must complete task objectives while following business rules. AutomationBench-AA uses Zapier’s private set of 657 tasks, and is built on v1.0.6 Key results: ➤ Claude Fable 5.1 and GPT-6 Astra lead the Intelligence Index: Both Claude Fable 5.1 (max with fallback) and GPT-6 Astra (max) score 53 on Intelligence Index v4.3, followed by Claude Opus 5 (max, 51), Claude Fable 5 (with fallback, 50), Muse Spark 1.3 (max, 48) and GPT-5.6 Sol (max, 47) ➤ GLM-5.3 and Kimi K3 continue to lead open weights models (both at 44): GLM-5.3-Flash (42) is the third strongest open weights model, followed by Qwen3.8 2.4T A95B (40) and DeepSeek V4 Pro 0813 (max, 36) ➤ 4 labs occupy the Intelligence vs. Cost per Task Pareto frontier: OpenAI occupies the majority of the cost-efficiency frontier, with all five reasoning efforts of the recently released GPT-6 Astra offering the lowest Cost per Task at their respective levels of intelligence. Claude Fable 5.1 (xhigh, max, 53), GLM-5.3-Flash (42) and MiMo-V2.5-Pro (26) round out the rest of the frontier

📊 RWA holder count crosses 1.3M. Looking at the data, the pace of new RWA holders accelerated after the 1M mark on July 9. Every prior 100K milestone took at least 15 days, with some taking months. 1M → 1.1M: 7 days 1.1M → 1.2M: 2 days 1.2M → 1.3M: 5 days

Nvidia dropped an official DeepSeek-V4.1-Flash NVFP4 build. And this thing is BIG. DeepSeek V4.1 Flash stats 🧠 552B backbone 📚 +196B Engram conditional memory ⚡ only 8B active during prefill 🚀 16B active during decode 👁️ native vision 📖 1 MILLION token context 🧩 384 routed experts across 40 layers 📜 MIT license Nvidia has converted its routed MoE experts to NVFP4 W4A4 specifically for Blackwell GPUs. This is not a compressed giant model down to 4-bit. DeepSeek's experts were already stored in MXFP4, Nvidia instead converts them to its Blackwell-friendly NVFP4 format. And because NVFP4 uses finer scaling, the checkpoint actually gets slightly larger. 💾 Source: ~476 GiB 💾 NVIDIA NVFP4: ~492 GiB 48 safetensor shards. 😳 So why bother? Because NVIDIA is optimizing how those 4-bit experts execute on Blackwell. And impressively, NVIDIA's evaluations show basically no obvious quality collapse from the conversion. For example: 🧠 GPQA Diamond 91.04 → 91.29 💻 SciCode 54.40 → 55.84 🛠️ Terminal-Bench 2.1 81.60 → 82.16 👁️ MMMU-Pro 74.05 → 73.70 Some slightly up. Some slightly down. Essentially benchmark parity. And it already has: ✅ vLLM support ✅ SGLang support ✅ reasoning parser ✅ tool calling ✅ image input ✅ 1M context ⚠️ Nvidia validated it on 4× GB300 GPUs so not a local model (yet). The checkpoint is still ~492 GiB. 🔗 HF: /nvidia/DeepSeek-V4.1-Flash-NVFP4

Let me give you a challenge. You don’t have to do much—just think of any number. If it’s even, divide it by 2. If it’s odd, apply the rule 3𝑛 + 1, where 𝑛 is the number you chose. Then take the result and repeat the same process again. If you keep doing these calculations, you’ll eventually notice that the sequence always ends at 1. And if you continue further, the numbers 4 → 2 → 1 will repeat over and over again. Now here’s the challenge: try this with as many numbers as you can think of. If you ever find a number that does not end in the 4 → 2 → 1 cycle, congratulations—you’ve just solved an 88-year-old mathematical mystery. Yes, mathematicians have tested trillions of numbers over the past 88 years, and so far, not a single number has been found that avoids ending in the 4 → 2 → 1 sequence. This is known as the 3𝑛 + 1 problem (Collatz conjecture).

Digit lost its bird legs! 🦿 @agilityrobotics released Digit 5 today, and the most visible change is the knees. The ostrich-style backwards legs are gone, replaced with human-like ones. Jonathan Hurst, co-founder and chief robot officer, was explicit that this wasn't a move toward looking more human: 💬 "The previous leg configuration was really optimized for very good walking and running. The new configuration is optimized for lifting heavy things and getting up and down off the ground a lot." The same logic shows up in the hands, where the argument gets I would say even sharper. He says general-purpose hands don't exist, aren't on the market, and aren't coming soon. His evidence is Cybathlon, where amputees compete at manipulation tasks using prosthetics. The winner two years running used a single-degree-of-freedom parallel-jaw gripper with one rotation at the wrist, beating competitors with five-fingered hands. So Digit 5 runs a two-finger gripper, swappable through ISO-standard mounting flanges: → Repeatable lifts up to 22.7 kg on proprietary cycloidal actuators, covering single-person lift tasks in OSHA-regulated facilities → 90 minutes of work, 9 minutes to full charge. A 10:1 run-to-charge ratio, up from 2:1 on Digit 4, which puts it past 20 working hours in a 24-hour day → 129 kg, 1.81 m tall, reaching to 2.2 m, up from 1.68 m → 360° person detection with no fixed safety zones. It slows, stops, or sits down and cuts motor power before anyone can touch it, with an independent safety controller overseeing the response First commercial availability outside North America, starting with the EU and UK, CE mark expected. Early access in H1 2027, general availability by the end of 2027. ~~  ♻️ Join the weekly robotics newsletter, and never miss any news →

Newly released v0.2.1 node manager introduces quantum jobs submissions to the network, and other features: ‣ One miner for all hardware types ‣ Expanded post-quantum signing ‣ More resilient nodes ‣ Improved on-chain transparency ‣ And more... →

Later today we'll adjust two parameters: Purchase surcharge: 2.5% → 1.5% Buyback allocation: 40/30/30 → 60/20/20 (purchasers/depositors/burn). We'll continue to monitor and adjust small mechanisms like this for the benefit of the protocol

🔥 Word is ByteDance's Seedance 2.1 is coming early July. What's reportedly landing: • Clip length 15s → 30s • Native 4K output • Major quality & performance jump There's also a lighter Seedance 2.0 mini on the way — cheaper, faster . We're barely done with 2.0 and 30s native 4K is already around the corner. AI video is moving absurdly fast. #Seedance# #AIvideo# #ByteDance#