DeepSeek-V4.1-Flash is now live on Nebius Token Factory The model comes with native image understanding plus an architecture built for faster inference and higher throughput on long, input-heavy workflows. It combines native vision and a 1M token context with a 552B MoE that activates just 8B parameters for input and 16B for output. Try it out: tokenfactory.nebius.com/?mod…

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

$PENG turned scarcity into revenue. Samsung made memory stretch. Micron locked down wafers. Korea broke the wrapper. this week, the book looked flat but the stack was being repriced ↓ Under a flat headline, the market made a clear choice: pay for what can turn scarcity into revenue now. Storage, networking, AI cloud and cooling led. HBM / packaging and power / grid lagged. $PENG, Samsung, Micron and Korea explain the rotation. REVENUE NOW - $PENG Penguin was the receipt. Q3 revenue grew 48%. Integrated Memory more than doubled. Operating income rose more than 5x. The point is not just that memory is tight. Penguin gets paid to turn memory, clusters and infrastructure software into a working AI system. Scarcity matters but making it usable pays sooner. MEMORY ELASTICITY - SAMSUNG Samsung's Blackwell test added a 1TB CXL pool for KV-cache offload. When local DRAM filled, throughput collapsed. The CXL-backed system kept running near DRAM speed (approximately 92% of DRAM performance in multi-GPU configurations). Not because CXL replaces HBM but because not every byte needs the fastest, most expensive tier. The hierarchy is widening: HBM → DRAM → pooled CXL → storage The caveat: this required host changes, a custom in-house kernel and modifications to the LMCache stack. Engineering proof, not plug-and-play adoption. But it shows where the next memory trade may form: deciding where each byte belongs. SUPPLY SECURITY - MICRON Micron put $500M behind GlobalWafers' 300mm Texas plant and signed a 10-year supply agreement. That is more revealing than another demand forecast. When the customer finances the supplier, the bottleneck has moved from the presentation deck to capital allocation. AI memory profits are being recycled into the layer beneath memory: wafers, materials and geographic resilience. RACK TIME VS GRID TIME Transformer queues now stretch beyond three years. High-voltage breakers are not far behind. Yet cooling rallied while Power & Grid fell. The physical constraint did not vanish. The market separated two clocks: Cooling gets installed with the rack. Grid equipment gets paid after permitting, financing and construction. Same density problem. Very different route to cash. DEPLOYABILITY BEAT DESTINY Networking and retimers outperformed photonics / CPO. $ANET led; $CRDO gained. That is not a verdict against optics. CPO may still be the architectural destination. But copper, retimers and systems already shipping can earn during the transition. The market paid for what can deploy before the perfect end state arrives. SAME ASSET, DIFFERENT PLUMBING SK hynix's U.S. ADR jumped on debut. The next trading day, its Seoul shares fell more than 15%, Samsung fell sharply, the KOSPI fell ~9% and trading halted. Reuters calculated a roughly 37% ADR premium after the rout. Same company, same HBM exposure, but different access, liquidity, leverage and flows. Korea did not prove that the HBM thesis was broken. It proved that the wrapper can overpower the asset. THE THREE CLOCKS 1. Revenue now: storage, networking, AI operations, cooling. 2. Scarcity later: power, fabs, packaging, substrates. 3. Elasticity: CXL, retimers, liquid cooling, orchestration. That is the map for this week: $PENG = turn scarcity into revenue. Samsung = make scarce memory go further. $MU = secure the layer underneath. Korea = price the instrument, not only the asset. The bottleneck did not disappear. The market started asking a harder question: How long until it becomes cash? Full weekly: aibottlenecks.app/alpha/week…

Artificial Analysis currently ranks Nebius first among 12 providers for GLM‑5.3‑Flash on: ⚡ Output speed: ~290 tokens/second ⏱️ End-to-end response time: ~9.1 seconds A good reminder that choosing the model is only half the job. The inference stack determines the experience users actually get. See the full comparison: artificialanalysis.ai/models…

Ox Alpha has a name: GLM-5.3-Flash. @Zai_org’s new open-source model is now live on Nebius Token Factory on Day 0, with a 1M-token context window, multimodal capabilities, and zero data retention. Build with it: tokenfactory.nebius.com/endp…

Ox Alpha has a name: GLM-5.3-Flash. @Zai_org’s new open-source model is now live on Nebius Token Factory on Day 0, with a 1M-token context window, multimodal capabilities, and zero data retention. Build with it: tokenfactory.nebius.com/endp…

OK Now officially Very soon at @nebiustf The team is on that:)