Technology News
886 articles
AI technology breakthroughs, research papers, patents, and technical innovations from leading AI startups.
WEKA pitches shared flash KV cache to reduce LLM inference recomputation
WEKA is pitching its Augmented Memory Grid, running on NeuralMesh storage software, as a way to reuse LLM context across GPU servers. The system captures KV cache and makes it available across a data...

PrismML adds three smaller models to its Ternary Bonsai 2 lineup
PrismML has released Qwen3.5-based Ternary Bonsai 2 models with 2B, 4B and 9B parameters, extending a lineup that already includes a 27B version. Their ternary weights encode parameters as {-1, 0, +1}...
Rebellions and FuriosaAI showcase server inference chips for physical AI
Rebellions and FuriosaAI displayed inference hardware at Physical AI Tech Day in Seoul on October 14, showing how servers can process inputs from robots. Rebellions exhibited Atom Max and RebelCard, w...

Ant Group releases Ling-3.1-flash weights with four-GPU deployment option
Ant Group has released the weights of Ling-3.1-flash for developers to deploy on their own infrastructure. The model has approximately 560 billion total parameters, activates about 25 billion per toke...

TypeSafe AI's Jev reaches 35% of paid teams on Vercel's AI gateway
TypeSafe AI's Jev was used by 35% of paid teams on Vercel's AI gateway within two weeks of launch, according to Vercel data reported by TechCrunch. Its reach was nearly twice that of runner-up Sonnet...
Reflection releases Beam open weights with a focus on using fewer tokens
Reflection released the weights for Beam, a 501-billion-parameter model with 23 billion active parameters per token. The company says reinforcement learning trained Beam to complete tasks with fewer t...

Nvidia RTX Spark pairs local AI hardware with Windows agent distribution
Microsoft has adapted Windows for Nvidia's RTX Spark laptops, combining an Arm CPU, Blackwell GPU, and unified memory with a new agent setup experience. According to The Verge, Windows offers a Get St...
NVIDIA releases Nemotron Super VL and Bolt for specialized AI workloads
NVIDIA released open-weight Nemotron models targeting document processing, vision-language tasks, and local agent workloads. Super VL addresses document and visual inputs, while the Bolt models target...
NVIDIA adds Nemotron 3.5 Super VL and Bolt to its open model family
NVIDIA has expanded its Nemotron open model family with Nemotron 3.5 Super VL and Nemotron 3.5 Bolt, targeting always-on agents and an additional deployment tier. The company claims 5x faster performa...
Perplexity Open-Sources SPACE for Sandboxed AI Agent Code Execution
Perplexity has released SPACE, the sandbox platform it uses to isolate code execution by its AI agents, as an open-source project. The platform is designed to keep agents within defined execution cons...

Ricursive Intelligence targets whole-chip optimization with AI
Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini outlined their approach to AI-driven chip design at TechCrunch Disrupt. The startup is developing systems that design chips, learn fro...

Skywork AI releases Matrix-Game 4.0 for AI-rendered gameplay on one RTX 5090
Kunlun Tech released Matrix-Game 4.0 on October 15, combining a game engine with real-time AI video rendering. The company says the system delivers 480p gameplay at 60 frames per second on a single Nv...
Third Moment Research reports running 10,000 AI agents in extended research sessions
Third Moment Research uses OpenAI models and Codex to run as many as 10,000 agents concurrently on quantitative research, with sessions lasting 36–48 hours. The company says Codex helped reduce a back...

Niobia uses equipment-mounted AI to diagnose chip and battery production defects
San Francisco-based Niobia is developing AI software to detect manufacturing defects and help engineers investigate their causes. Its agents run on self-contained GPUs mounted directly on production e...
NVIDIA Open-Sources Full Run:ai GPU Orchestration Platform
NVIDIA has made the full Run:ai platform available as open-source software. The platform helps customers and developers manage GPU infrastructure and workloads. The announcement concerns the software...

Flox Intelligence deploys adaptive edge AI to deter wildlife near infrastructure
Flox Intelligence, a KTH spinout founded in 2020, uses on-device animal recognition and species-appropriate sounds to deter wildlife near transport and infrastructure. Motion sensors activate cameras,...

Microsoft plans multigigawatt Maia deployment from 2027 with inference-focused Maia 300
Microsoft plans to expand its Azure Maia accelerator fleet to multigigawatt scale from 2027 and make it a primary inference platform for its products and services. Its third-generation Maia 300 target...

Anthropic Red Team Warns of Potential Self-Replicating AI Threats
Anthropic's Frontier Red Team suggests some AI models may be months away from capabilities sufficient for self-replication, according to The Verge. Its report examines scenarios in which small models...

Microsoft announces Maia 300, targeting primary inference role in 2027
Microsoft announced Maia 300, its third-generation in-house AI accelerator, with production rollout planned for 2027. Microsoft expects 40% more tokens per megawatt than leading GPU-based systems and...

Meta unveils Muse Wonder models with persistent memory and split API access
Meta Superintelligence Labs has introduced Muse Wonder and Muse Wonder Full, larger models than its lightweight Muse Spark, according to ZDNet Korea. Meta says they improve coding and complex task han...