IR4 research
Technical articles
42 articles · 22 companies · updated 24 August 202642 articles
- 064Google AI Developer and Agent PlatformHow AI Studio, Antigravity, Gemini CLI, Code Assist, Jules, ADK and Agent Platform fit together.GoogleAgent system
- 063Alphabet & Google AI: Organisations, Models, Products and InfrastructureHow Alphabet, Google, Google DeepMind, Cloud, product teams, models and infrastructure fit together.GoogleCompany dossier
- 062Codex: Agent Architecture, Tools and Product SystemHow models, context, tools, permissions, skills, plugins, MCP, subagents, cloud execution and developer interfaces form one agent system.OpenAIAgent system
- 061OpenAI: Company, Models and Product SystemHow governance, compute, model families, ChatGPT, Codex and the developer platform fit together.OpenAICompany dossier
- 060Claude Code: Agent Architecture, Tools and Product SystemHow models, context, tools, Skills, MCP, subagents, the Agent SDK and security controls form one agent system.AnthropicAgent system
- 059Anthropic: Company, Models and Product SystemHow governance, compute, model generations, direct products and the developer platform fit together.AnthropicCompany dossier
- 058AST SpaceMobile: Network, Partners and Business ModelHow giant phased arrays, spectrum, gateways and mobile operators are intended to connect ordinary phones.AST SpaceMobileCompany dossier
- 057Rocket Lab: Launch and End-to-End Space SystemsHow launch vehicles, spacecraft, components, software, acquisitions and the proposed Iridium transaction fit together.Rocket LabCompany dossier
- 056Blue Origin: Launch, Lunar and Orbital InfrastructureHow reusable transport, engines, lunar systems, orbital mobility, networks and destinations fit together.Blue OriginCompany dossier
- 055Starlink: How the Network Works and Where It Can ExpandSatellites, laser links, terminals and the customer path across air, sea, road, rail, cargo and mobile.SpaceXNetwork architecture
- 054SpaceX: Business Structure, Technology and EconomicsHow Space, Connectivity and AI fit together—including xAI, X, Cursor, customers, control and roadmap.SpaceXCompany dossier
- 053CursorThe editor, coding-agent harness, cloud execution and automation surfaces now within SpaceXAI.SpaceXAI / CursorDevelopment platform
- 049Grok BuildThe open-source coding-agent harness, its tool loop, parallel subagents and extension system.xAICoding agent
- 036Qwen3-Max-ThinkingAdaptive tool use and experience-cumulative test-time scaling in Qwen’s hosted reasoner.QwenReasoning
- 035Qwen3.6Hybrid DeltaNet-attention MoE design, 3B active parameters and multimodal agentic coding.QwenOpen model
- 034Kimi K3How a 2.8T sparse multimodal model combines KDA, Gated MLA, AttnRes and LatentMoE.Moonshot AIOpen model
- 033Phi-4-reasoningA 14B model shaped by curated reasoning demonstrations and outcome-based reinforcement learning.MicrosoftOpen model
- 032Llama 4 BehemothWhat Meta disclosed about its nearly 2T teacher model—and why it remains unreleased.MetaPreview
- 031Olmo HybridWhy a 3:1 recurrent-attention pattern can combine state tracking with precise recall.Ai2Open research
- 030DeepSeek-V4-ProToken compression, sparse attention and 49B active parameters inside a 1.6T model.DeepSeekOpen model
- 029Cosmos 3A two-tower physical-AI model joining multimodal reasoning with world and action generation.NVIDIAPhysical AI
- 028Nemotron 3 UltraHybrid Mamba–Transformer reasoning, sparse experts and NVFP4 serving for long agents.NVIDIAOpen model
- 027Grok Imagine Video 1.5Text- and image-conditioned temporal generation with video and synchronised audio output.xAIVideo model
- 026Grok Imagine Image 2.0Text-to-image generation and image-conditioned editing at 1K and 2K.xAIImage model
- 025Grok 4.6A 500K-context reasoning model with image input, configurable effort and optional live search.xAIReasoning
- 017Intel 14ARibbonFET 2, PowerDirect, High-NA EUV, Turbo Cells and the node’s conditional development status.IntelFoundry
- 016TSMC A13 Logic TechnologyHow a 97% optical shrink extends A14 with 6% die-area saving and compatible design rules.TSMCFoundry
- 015Snapdragon C PlatformWhat is confirmed—and still undisclosed—about Qualcomm’s entry-tier Windows laptop processor.QualcommPlatform
- 014Dragonfly C1000 Data-centre CPUHow 250-plus Oryon cores, chiplets and PCIe 7 are designed to host cloud and AI systems.QualcommCompute
- 013Snapdragon X2 PlusTwo 3 nm PC processors with Oryon CPU cores, an 80-TOPS NPU and 152 GB/s memory.QualcommCompute
- 012Cortex AI Training ClustersTesla’s disclosed training capacity, operational model and still-unpublished system details.TeslaSystems
- 011Google Tensor Processing UnitsWhy TPU 8t and TPU 8i use different compute, memory and interconnect designs.GoogleCompute
- 010Xeon 6+ E-core / Clearwater ForestUp to 288 E-cores on Intel 18A with 3D chiplets and a twelve-channel memory system.IntelCompute
- 009Arm AGI CPUHow 136 Neoverse V3 cores, DDR5 and PCIe 6 feed and orchestrate AI systems.ArmCompute
- 008Apple M5 familyCPU, GPU, Neural Engine, unified memory and the two-die Pro/Max architecture.AppleCompute
- 007Spartan 7 FPGAsHow a 28 nm FPGA becomes custom parallel logic, memory, arithmetic and I/O.AMDFPGAs
- 006Instinct MI455XHow eight CDNA 5 compute chiplets, 432 GB HBM4 and scale-up links form an AI accelerator.AMDCompute
- 005Ryzen 9000 SeriesZen 5 desktop chiplets, AM5 I/O and the current single- and dual-CCD 3D V-Cache designs.AMDCompute
- 004Ryzen AI Max 300 / Strix HaloHow CPU chiplets, a large integrated GPU, an NPU and 256-bit unified memory work together.AMDCompute
- 003ConnectX-9 SuperNICHow an 800 Gb/s endpoint connects GPU servers to Ethernet or InfiniBand scale-out fabrics.NVIDIANetworking
- 002Vera Rubin GPU PlatformHow seven chip types and five rack-scale systems are co-designed as one AI factory platform.NVIDIAPlatform
- 001Vera Rubin NVL72What the 72-GPU rack is and how its processors, memory and interconnects work together.NVIDIASystems
No articles match these filters.